diff --git a/cngram2vec.c b/cngram2vec.c
new file mode 100644
index 0000000..266cf76
--- /dev/null
+++ b/cngram2vec.c
@@ -0,0 +1,1442 @@
+//  Copyright 2013 Google Inc. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <pthread.h>
+
+#define MAX_STRING 100
+#define EXP_TABLE_SIZE 1000
+#define MAX_EXP 6
+#define MAX_SENTENCE_LENGTH 1000
+#define MAX_CODE_LENGTH 40
+
+const int vocab_hash_size = 30000000;  // Maximum 30 * 0.7 = 21M words in the vocabulary
+
+typedef float real;                    // Precision of float numbers
+
+struct vocab_word {
+  long long cn;
+  int *point;
+  char *word, *code, codelen;
+};
+
+char train_file[MAX_STRING], output_file[MAX_STRING];
+char save_vocab_file[MAX_STRING], read_vocab_file[MAX_STRING];
+struct vocab_word *vocab;
+int binary = 0, type = 1, debug_mode = 2, window = 5, min_count = 5, num_threads = 12, min_reduce = 1;
+int *vocab_hash;
+long long vocab_max_size = 1000, vocab_size = 0, layer1_size = 100;
+long long train_words = 0, word_count_actual = 0, iter = 5, file_size = 0, classes = 0;
+real alpha = 0.025, starting_alpha, sample = 1e-3;
+real *syn0, *syn1, *syn1neg, *syn1nce, *expTable;
+clock_t start;
+
+real *syn1_window, *syn1neg_window, *syn1nce_window;
+int w_offset, window_layer_size;
+
+int window_hidden_size = 500; 
+real *syn_window_hidden, *syn_hidden_word, *syn_hidden_word_neg, *syn_hidden_word_nce; 
+
+int hs = 0, negative = 5;
+const int table_size = 1e8;
+int *table;
+
+//constrastive negative sampling
+char negative_classes_file[MAX_STRING];
+int *word_to_group;
+int *group_to_table; //group_size*table_size
+int class_number;
+
+//nce
+real* noise_distribution;
+int nce = 10;
+
+//param caps
+real CAP_VALUE = 50;
+int cap = 0;
+
+// char models
+char boundToken = 'Z';
+char *unkNgramToken = "ZZZ";
+int cngram_size = 6;
+real *syn0_cngram;
+long long cngram_vocab_size = 0;
+struct vocab_word *cngram_vocab;
+int *cngram_vocab_hash;
+long long cngram_vocab_max_size = 1000;
+char extra_vocab_file[MAX_STRING];
+long long maxNgramSize = 1000000;
+
+// Returns hash value of a word
+int GetWordHash(char *word) {
+  unsigned long long a, hash = 0;
+  for (a = 0; a < strlen(word); a++) hash = hash * 257 + word[a];
+  hash = hash % vocab_hash_size;
+  return hash;
+}
+
+// Search
+int SearchCNgramVocab(char *ngram) {
+  unsigned int hash = GetWordHash(ngram);
+  while (1) {
+    if (cngram_vocab_hash[hash] == -1) return -1;
+    if (!strcmp(ngram, cngram_vocab[cngram_vocab_hash[hash]].word)) return cngram_vocab_hash[hash];
+    hash = (hash + 1) % vocab_hash_size;
+  }
+  return -1;
+}
+
+// char functions
+void ForwardCNgramWordNgram(real *output, char *ngram){
+	long long a;
+	int index = SearchCNgramVocab(ngram);
+	if (index == -1) {index = SearchCNgramVocab(unkNgramToken);}
+	long long startIndex = layer1_size * index;
+	for (a = 0; a < layer1_size; a++){
+		output[a] += syn0_cngram[startIndex + a];
+	}
+}
+
+void ForwardCNgramWordRepresentation(real *output, char *word){
+        int length = strlen(word);
+        int start;
+        int cur_len;
+        char *ngram;
+        char tmp[cngram_size+1];
+        tmp[cngram_size] = '\0';
+	int ngrams = 0;
+        for(start = 0; start < length-cngram_size+1; start++){
+                ngram = word + start;
+                strncpy(tmp, ngram, cngram_size);
+                ForwardCNgramWordNgram(output, tmp);
+		ngrams++;
+        }
+	for(cur_len = 0; cur_len < cngram_size-1; cur_len++) tmp[cur_len] = boundToken;
+        strncpy(tmp+1, word, cur_len);
+        ForwardCNgramWordNgram(output, tmp);
+        for(cur_len = 0; cur_len < cngram_size-1; cur_len++) tmp[cngram_size-cur_len-1] = boundToken;
+        cur_len = cngram_size - 1;
+        if(length < cur_len){
+                cur_len = length;
+        }
+        ngram = word + length - cur_len;
+        strncpy(tmp, ngram, cur_len);
+        tmp[cur_len] = 'Z';
+        tmp[cur_len + 1] = '\0';
+        ForwardCNgramWordNgram(output, tmp);
+	for(start = 0; start < layer1_size; start++){
+		output[start] /= ngrams+2;
+	}
+}
+
+void BackwardCNgramWordNgram(real *output, char *ngram, real *output_err){
+        long long a;
+        int index = SearchCNgramVocab(ngram);
+        if (index == -1) index = SearchCNgramVocab(unkNgramToken);
+        long long startIndex = layer1_size * index;
+        for (a = 0; a < layer1_size; a++){
+                syn0_cngram[startIndex + a] += output_err[a];
+        }
+}
+
+void BackwardCNgramWordRepresentation(real *output, char *word, real *output_err){
+	int length = strlen(word);
+        int start;
+        int cur_len;
+        char *ngram;
+        char tmp[cngram_size+1];
+        tmp[cngram_size] = '\0';
+        for(start = 0; start < length-cngram_size+1; start++){
+                ngram = word + start;
+                strncpy(tmp, ngram, cngram_size);
+                BackwardCNgramWordNgram(output, tmp, output_err);
+        }
+	for(cur_len = 0; cur_len < cngram_size-1; cur_len++) tmp[cur_len] = boundToken;
+        strncpy(tmp+1, word, cur_len);
+        BackwardCNgramWordNgram(output, tmp, output_err);
+        for(cur_len = 0; cur_len < cngram_size-1; cur_len++) tmp[cngram_size-cur_len-1] = boundToken;
+        cur_len = cngram_size - 1;
+        if(length < cur_len){
+                cur_len = length;
+        }
+        ngram = word + length - cur_len;
+        strncpy(tmp, ngram, cur_len);
+        tmp[cur_len] = 'Z';
+        tmp[cur_len + 1] = '\0';
+        BackwardCNgramWordNgram(output, tmp, output_err);
+}
+
+void AddWordNgramToVocab(char *ngram, int count){
+	int index = SearchCNgramVocab(ngram);
+	if(index != -1){
+		cngram_vocab[index].cn+=count;
+		return;
+	}
+        unsigned int hash, length = strlen(ngram) + 1;
+        if (length > MAX_STRING) length = MAX_STRING;
+        cngram_vocab[cngram_vocab_size].word = (char *)calloc(length, sizeof(char));
+        strcpy(cngram_vocab[cngram_vocab_size].word, ngram);
+        cngram_vocab[cngram_vocab_size].cn = count;
+        cngram_vocab_size++;
+        // Reallocate memory if needed
+        if (cngram_vocab_size + 2 >= cngram_vocab_max_size) {
+                cngram_vocab_max_size += 1000;
+        	cngram_vocab = (struct vocab_word *)realloc(cngram_vocab, cngram_vocab_max_size * sizeof(struct vocab_word));
+        }
+        hash = GetWordHash(ngram);
+        while (cngram_vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+        cngram_vocab_hash[hash] = cngram_vocab_size - 1;
+}
+
+void AddAllWordNgramToVocab(char *word, int count){
+	int length = strlen(word);
+	int start;
+	int cur_len;
+	char *ngram;
+	char tmp[cngram_size+1];
+	tmp[cngram_size] = '\0';
+	for(start = 0; start < length-cngram_size+1; start++){
+		ngram = word + start;
+		strncpy(tmp, ngram, cngram_size);
+		AddWordNgramToVocab(tmp, count);
+	}
+	for(cur_len = 0; cur_len < cngram_size-1; cur_len++) tmp[cur_len] = boundToken;
+	strncpy(tmp+1, word, cur_len);
+	AddWordNgramToVocab(tmp, count);
+	for(cur_len = 0; cur_len < cngram_size-1; cur_len++) tmp[cngram_size-cur_len-1] = boundToken;
+	cur_len = cngram_size - 1;
+	if(length < cur_len){
+		cur_len = length;
+	}
+	ngram = word + length - cur_len;
+	strncpy(tmp, ngram, cur_len);
+	tmp[cur_len] = 'Z';
+	tmp[cur_len + 1] = '\0';
+	AddWordNgramToVocab(tmp, count);	
+}
+
+void capParam(real* array, int index){
+	if(array[index] > CAP_VALUE) 
+		array[index] = CAP_VALUE;
+	else if(array[index] < -CAP_VALUE)
+		array[index] = -CAP_VALUE; 
+}
+
+real hardTanh(real x){
+	if(x>=1){
+		return 1;
+	}
+	else if(x<=-1){
+		return -1;
+	}
+	else{
+		return x;
+	}
+}
+
+real dHardTanh(real x, real g){
+	if(x > 1 && g > 0){
+		return 0;
+	}
+	if(x < -1 && g < 0){
+		return 0;
+	}
+	return 1;
+}
+
+void InitUnigramTable() {
+  int a, i;
+  long long train_words_pow = 0;
+  real d1, power = 0.75;
+  table = (int *)malloc(table_size * sizeof(int));
+  for (a = 0; a < vocab_size; a++) train_words_pow += pow(vocab[a].cn, power);
+  i = 0;
+  d1 = pow(vocab[i].cn, power) / (real)train_words_pow;
+  for (a = 0; a < table_size; a++) {
+    table[a] = i;
+    if (a / (real)table_size > d1) {
+      i++;
+      d1 += pow(vocab[i].cn, power) / (real)train_words_pow;
+    }
+    if (i >= vocab_size) i = vocab_size - 1;
+  }
+  
+  noise_distribution = (real *)calloc(vocab_size, sizeof(real));
+  for (a = 0; a < vocab_size; a++) noise_distribution[a] = pow(vocab[a].cn, power)/(real)train_words_pow;
+}
+
+// Reads a single word from a file, assuming space + tab + EOL to be word boundaries
+void ReadWord(char *word, FILE *fin) {
+  int a = 0, ch;
+  while (!feof(fin)) {
+    ch = fgetc(fin);
+    if (ch == 13) continue;
+    if ((ch == ' ') || (ch == '\t') || (ch == '\n')) {
+      if (a > 0) {
+        if (ch == '\n') ungetc(ch, fin);
+        break;
+      }
+      if (ch == '\n') {
+        strcpy(word, (char *)"</s>");
+        return;
+      } else continue;
+    }
+    word[a] = ch;
+    a++;
+    if (a >= MAX_STRING - 1) a--;   // Truncate too long words
+  }
+  word[a] = 0;
+}
+
+// Returns position of a word in the vocabulary; if the word is not found, returns -1
+int SearchVocab(char *word) {
+  unsigned int hash = GetWordHash(word);
+  while (1) {
+    if (vocab_hash[hash] == -1) return -1;
+    if (!strcmp(word, vocab[vocab_hash[hash]].word)) return vocab_hash[hash];
+    hash = (hash + 1) % vocab_hash_size;
+  }
+  return -1;
+}
+
+// Reads a word and returns its index in the vocabulary
+int ReadWordIndex(FILE *fin) {
+  char word[MAX_STRING];
+  ReadWord(word, fin);
+  if (feof(fin)) return -1;
+  return SearchVocab(word);
+}
+
+// Adds a word to the vocabulary
+int AddWordToVocab(char *word) {
+  unsigned int hash, length = strlen(word) + 1;
+  if (length > MAX_STRING) length = MAX_STRING;
+  vocab[vocab_size].word = (char *)calloc(length, sizeof(char));
+  strcpy(vocab[vocab_size].word, word);
+  vocab[vocab_size].cn = 0;
+  vocab_size++;
+  // Reallocate memory if needed
+  if (vocab_size + 2 >= vocab_max_size) {
+    vocab_max_size += 1000;
+    vocab = (struct vocab_word *)realloc(vocab, vocab_max_size * sizeof(struct vocab_word));
+  }
+  hash = GetWordHash(word);
+  while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+  vocab_hash[hash] = vocab_size - 1;
+  return vocab_size - 1;
+}
+
+// Used later for sorting by word counts
+int VocabCompare(const void *a, const void *b) {
+    return ((struct vocab_word *)b)->cn - ((struct vocab_word *)a)->cn;
+}
+
+// Sorts the vocabulary by frequency using word counts
+void SortVocab() {
+  int a, size;
+  unsigned int hash;
+  // Sort the vocabulary and keep </s> at the first position
+  qsort(&vocab[1], vocab_size - 1, sizeof(struct vocab_word), VocabCompare);
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  size = vocab_size;
+  train_words = 0;
+  for (a = 0; a < size; a++) {
+    // Words occuring less than min_count times will be discarded from the vocab
+    if ((vocab[a].cn < min_count) && (a != 0)) {
+      vocab_size--;
+      free(vocab[a].word);
+    } else {
+      // Hash will be re-computed, as after the sorting it is not actual
+      hash=GetWordHash(vocab[a].word);
+      while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+      vocab_hash[hash] = a;
+      train_words += vocab[a].cn;
+    }
+  }
+  vocab = (struct vocab_word *)realloc(vocab, (vocab_size + 1) * sizeof(struct vocab_word));
+  // Allocate memory for the binary tree construction
+  for (a = 0; a < vocab_size; a++) {
+    vocab[a].code = (char *)calloc(MAX_CODE_LENGTH, sizeof(char));
+    vocab[a].point = (int *)calloc(MAX_CODE_LENGTH, sizeof(int));
+  }
+}
+
+// Reduces the vocabulary by removing infrequent tokens
+void ReduceVocab() {
+  int a, b = 0;
+  unsigned int hash;
+  for (a = 0; a < vocab_size; a++) if (vocab[a].cn > min_reduce) {
+    vocab[b].cn = vocab[a].cn;
+    vocab[b].word = vocab[a].word;
+    b++;
+  } else free(vocab[a].word);
+  vocab_size = b;
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  for (a = 0; a < vocab_size; a++) {
+    // Hash will be re-computed, as it is not actual
+    hash = GetWordHash(vocab[a].word);
+    while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+    vocab_hash[hash] = a;
+  }
+  fflush(stdout);
+  min_reduce++;
+}
+
+// Create binary Huffman tree using the word counts
+// Frequent words will have short uniqe binary codes
+void CreateBinaryTree() {
+  long long a, b, i, min1i, min2i, pos1, pos2, point[MAX_CODE_LENGTH];
+  char code[MAX_CODE_LENGTH];
+  long long *count = (long long *)calloc(vocab_size * 2 + 1, sizeof(long long));
+  long long *binary = (long long *)calloc(vocab_size * 2 + 1, sizeof(long long));
+  long long *parent_node = (long long *)calloc(vocab_size * 2 + 1, sizeof(long long));
+  for (a = 0; a < vocab_size; a++) count[a] = vocab[a].cn;
+  for (a = vocab_size; a < vocab_size * 2; a++) count[a] = 1e15;
+  pos1 = vocab_size - 1;
+  pos2 = vocab_size;
+  // Following algorithm constructs the Huffman tree by adding one node at a time
+  for (a = 0; a < vocab_size - 1; a++) {
+    // First, find two smallest nodes 'min1, min2'
+    if (pos1 >= 0) {
+      if (count[pos1] < count[pos2]) {
+        min1i = pos1;
+        pos1--;
+      } else {
+        min1i = pos2;
+        pos2++;
+      }
+    } else {
+      min1i = pos2;
+      pos2++;
+    }
+    if (pos1 >= 0) {
+      if (count[pos1] < count[pos2]) {
+        min2i = pos1;
+        pos1--;
+      } else {
+        min2i = pos2;
+        pos2++;
+      }
+    } else {
+      min2i = pos2;
+      pos2++;
+    }
+    count[vocab_size + a] = count[min1i] + count[min2i];
+    parent_node[min1i] = vocab_size + a;
+    parent_node[min2i] = vocab_size + a;
+    binary[min2i] = 1;
+  }
+  // Now assign binary code to each vocabulary word
+  for (a = 0; a < vocab_size; a++) {
+    b = a;
+    i = 0;
+    while (1) {
+      code[i] = binary[b];
+      point[i] = b;
+      i++;
+      b = parent_node[b];
+      if (b == vocab_size * 2 - 2) break;
+    }
+    vocab[a].codelen = i;
+    vocab[a].point[0] = vocab_size - 2;
+    for (b = 0; b < i; b++) {
+      vocab[a].code[i - b - 1] = code[b];
+      vocab[a].point[i - b] = point[b] - vocab_size;
+    }
+  }
+  free(count);
+  free(binary);
+  free(parent_node);
+}
+
+void LearnVocabFromTrainFile() {
+  char word[MAX_STRING];
+  FILE *fin;
+  long long a, i;
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  for (a = 0; a < vocab_hash_size; a++) cngram_vocab_hash[a] = -1;
+  fin = fopen(train_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: training data file not found!\n");
+    exit(1);
+  }
+  vocab_size = 0;
+  AddWordToVocab((char *)"</s>");
+  AddWordNgramToVocab(unkNgramToken,1000000);
+  while (1) {
+    ReadWord(word, fin);
+    if (feof(fin)) break;
+    train_words++;
+    if ((debug_mode > 1) && (train_words % 100000 == 0)) {
+      printf("%lldK%c", train_words / 1000, 13);
+      fflush(stdout);
+    }
+    i = SearchVocab(word);
+    if (i == -1) {
+      a = AddWordToVocab(word);
+      vocab[a].cn = 1;
+    } else vocab[i].cn++;
+    if (vocab_size > vocab_hash_size * 0.7) ReduceVocab();
+  }
+  SortVocab();
+  for (a = 0; a < vocab_size; a++){
+      AddAllWordNgramToVocab(vocab[a].word, vocab[a].cn);
+  }
+  if (debug_mode > 0) {
+    printf("Vocab size: %lld\n", vocab_size);
+    printf("Ngrams size: %lld\n", cngram_vocab_size);
+    printf("Words in train file: %lld\n", train_words);
+  }
+  file_size = ftell(fin);
+  fclose(fin);
+}
+
+void SaveVocab() {
+  long long i;
+  FILE *fo = fopen(save_vocab_file, "wb");
+  for (i = 0; i < vocab_size; i++) fprintf(fo, "%s %lld\n", vocab[i].word, vocab[i].cn);
+  fclose(fo);
+}
+
+void ReadVocab() {
+  long long a, i = 0;
+  char c;
+  char word[MAX_STRING];
+  FILE *fin = fopen(read_vocab_file, "rb");
+  if (fin == NULL) {
+    printf("Vocabulary file not found\n");
+    exit(1);
+  }
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  vocab_size = 0;
+  while (1) {
+    ReadWord(word, fin);
+    if (feof(fin)) break;
+    a = AddWordToVocab(word);
+    fscanf(fin, "%lld%c", &vocab[a].cn, &c);
+    i++;
+  }
+  SortVocab();
+  if (debug_mode > 0) {
+    printf("Vocab size: %lld\n", vocab_size);
+    printf("Words in train file: %lld\n", train_words);
+  }
+  fin = fopen(train_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: training data file not found!\n");
+    exit(1);
+  }
+  fseek(fin, 0, SEEK_END);
+  file_size = ftell(fin);
+  fclose(fin);
+}
+
+void InitClassUnigramTable() {
+  long long a,c;
+  printf("loading class unigrams \n");
+  FILE *fin = fopen(negative_classes_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: class file not found!\n");
+    exit(1);
+  }
+  word_to_group = (int *)malloc(vocab_size * sizeof(int));
+  for(a = 0; a < vocab_size; a++) word_to_group[a] = -1;
+  char class[MAX_STRING];
+  char prev_class[MAX_STRING];
+  prev_class[0] = 0;
+  char word[MAX_STRING];
+  class_number = -1;
+  while (1) {
+    if (feof(fin)) break;
+    ReadWord(class, fin);
+    ReadWord(word, fin);
+    int word_index = SearchVocab(word);
+    if (word_index != -1){
+       if(strcmp(class, prev_class) != 0){
+	    class_number++;
+	    strcpy(prev_class, class);
+       }
+       word_to_group[word_index] = class_number;
+    }
+    ReadWord(word, fin);
+  }
+  class_number++;
+  fclose(fin);
+  
+  group_to_table = (int *)malloc(table_size * class_number * sizeof(int)); 
+  long long train_words_pow = 0;
+  real d1, power = 0.75;
+  
+  for(c = 0; c < class_number; c++){
+     long long offset = c * table_size;
+     train_words_pow = 0;
+     for (a = 0; a < vocab_size; a++) if(word_to_group[a] == c) train_words_pow += pow(vocab[a].cn, power);
+     int i = 0;
+     while(word_to_group[i]!=c && i < vocab_size) i++;
+     d1 = pow(vocab[i].cn, power) / (real)train_words_pow;
+     for (a = 0; a < table_size; a++) {
+	//printf("index %lld , word %d\n", a, i);
+	group_to_table[offset + a] = i;
+        if (a / (real)table_size > d1) {
+	   i++;
+           while(word_to_group[i]!=c && i < vocab_size) i++;
+	   d1 += pow(vocab[i].cn, power) / (real)train_words_pow;
+        }
+        if (i >= vocab_size) while(word_to_group[i]!=c && i >= 0) i--;
+     }
+  }
+}
+
+void InitNet() {
+  long long a, b;
+  unsigned long long next_random = 1;
+  window_layer_size = layer1_size*window*2;
+  a = posix_memalign((void **)&syn0, 128, (long long)vocab_size * layer1_size * sizeof(real));
+  if (syn0 == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  a = posix_memalign((void **)&syn0_cngram, 128, (long long)vocab_size * layer1_size * sizeof(real));
+  if (syn0_cngram == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  
+  if (hs) {
+    a = posix_memalign((void **)&syn1, 128, (long long)vocab_size * layer1_size * sizeof(real));
+    if (syn1 == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn1_window, 128, (long long)vocab_size * window_layer_size * sizeof(real));
+    if (syn1_window == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn_hidden_word, 128, (long long)vocab_size * window_hidden_size * sizeof(real));
+    if (syn_hidden_word == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++)
+     syn1[a * layer1_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_layer_size; b++)
+     syn1_window[a * window_layer_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_hidden_size; b++)
+     syn_hidden_word[a * window_hidden_size + b] = 0;
+  }
+  if (negative>0) {
+    a = posix_memalign((void **)&syn1neg, 128, (long long)vocab_size * layer1_size * sizeof(real));
+    if (syn1neg == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn1neg_window, 128, (long long)vocab_size * window_layer_size * sizeof(real));
+    if (syn1neg_window == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn_hidden_word_neg, 128, (long long)vocab_size * window_hidden_size * sizeof(real));
+    if (syn_hidden_word_neg == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++)
+     syn1neg[a * layer1_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_layer_size; b++)
+     syn1neg_window[a * window_layer_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_hidden_size; b++)
+     syn_hidden_word_neg[a * window_hidden_size + b] = 0;
+  }
+  if (nce>0) {
+    a = posix_memalign((void **)&syn1nce, 128, (long long)vocab_size * layer1_size * sizeof(real));
+    if (syn1nce == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn1nce_window, 128, (long long)vocab_size * window_layer_size * sizeof(real));
+    if (syn1nce_window == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn_hidden_word_nce, 128, (long long)vocab_size * window_hidden_size * sizeof(real));
+    if (syn_hidden_word_nce == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++)
+     syn1nce[a * layer1_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_layer_size; b++)
+     syn1nce_window[a * window_layer_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_hidden_size; b++)
+     syn_hidden_word_nce[a * window_hidden_size + b] = 0;
+  }
+  for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++) {
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    syn0[a * layer1_size + b] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / layer1_size;
+  }
+
+  for (a = 0; a < cngram_vocab_size; a++) for (b = 0; b < layer1_size; b++){
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    syn0_cngram[a * layer1_size + b] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / layer1_size;
+  }
+
+  a = posix_memalign((void **)&syn_window_hidden, 128, window_hidden_size * window_layer_size * sizeof(real));
+  if (syn_window_hidden == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  for (a = 0; a < window_hidden_size * window_layer_size; a++){
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    syn_window_hidden[a] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / (window_hidden_size*window_layer_size);
+  }
+
+  CreateBinaryTree();
+}
+
+void *TrainModelThread(void *id) {
+  long long a, b, d, cw, word, last_word, sentence_length = 0, sentence_position = 0;
+  long long word_count = 0, last_word_count = 0, sen[MAX_SENTENCE_LENGTH + 1];
+  long long l1, l2, c, target, label, local_iter = iter;
+  unsigned long long next_random = (long long)id;
+  real f, g;
+  clock_t now;
+  int input_len_1 = layer1_size;
+  int window_offset = -1;
+  if(type == 2 || type == 4){
+     input_len_1=window_layer_size;
+  }
+  real *neu1 = (real *)calloc(input_len_1, sizeof(real));
+  real *neu1e = (real *)calloc(input_len_1, sizeof(real));
+
+  int input_len_2 = 0;
+  if(type == 4){
+     input_len_2 = window_hidden_size;
+  }
+  real *neu2 = (real *)calloc(input_len_2, sizeof(real));
+  real *neu2e = (real *)calloc(input_len_2, sizeof(real));
+
+  FILE *fi = fopen(train_file, "rb");
+  fseek(fi, file_size / (long long)num_threads * (long long)id, SEEK_SET);
+  while (1) {
+    if (word_count - last_word_count > 10000) {
+      word_count_actual += word_count - last_word_count;
+      last_word_count = word_count;
+      if ((debug_mode > 1)) {
+        now=clock();
+        printf("%cAlpha: %f  Progress: %.2f%%  Words/thread/sec: %.2fk  ", 13, alpha,
+         word_count_actual / (real)(iter * train_words + 1) * 100,
+         word_count_actual / ((real)(now - start + 1) / (real)CLOCKS_PER_SEC * 1000));
+        fflush(stdout);
+      }
+      alpha = starting_alpha * (1 - word_count_actual / (real)(iter * train_words + 1));
+      if (alpha < starting_alpha * 0.0001) alpha = starting_alpha * 0.0001;
+    }
+    if (sentence_length == 0) {
+      while (1) {
+        word = ReadWordIndex(fi);
+        if (feof(fi)) break;
+        if (word == -1) continue;
+        word_count++;
+        if (word == 0) break;
+        // The subsampling randomly discards frequent words while keeping the ranking same
+        if (sample > 0) {
+          real ran = (sqrt(vocab[word].cn / (sample * train_words)) + 1) * (sample * train_words) / vocab[word].cn;
+          next_random = next_random * (unsigned long long)25214903917 + 11;
+          if (ran < (next_random & 0xFFFF) / (real)65536) continue;
+        }
+        sen[sentence_length] = word;
+        sentence_length++;
+        if (sentence_length >= MAX_SENTENCE_LENGTH) break;
+      }
+      sentence_position = 0;
+    }
+    if (feof(fi) || (word_count > train_words / num_threads)) {
+      word_count_actual += word_count - last_word_count;
+      local_iter--;
+      if (local_iter == 0) break;
+      word_count = 0;
+      last_word_count = 0;
+      sentence_length = 0;
+      fseek(fi, file_size / (long long)num_threads * (long long)id, SEEK_SET);
+      continue;
+    }
+    word = sen[sentence_position];
+    if (word == -1) continue;
+    for (c = 0; c < input_len_1; c++) neu1[c] = 0;
+    for (c = 0; c < input_len_1; c++) neu1e[c] = 0;
+    for (c = 0; c < input_len_2; c++) neu2[c] = 0;
+    for (c = 0; c < input_len_2; c++) neu2e[c] = 0;
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    b = next_random % window;
+    if (type == 0) {  //train the cbow architecture
+      // in -> hidden
+      cw = 0;
+      for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+	ForwardCNgramWordRepresentation(neu1, vocab[last_word].word);
+        cw++;
+      }
+      if (cw) {
+        for (c = 0; c < layer1_size; c++) neu1[c] /= cw;
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * layer1_size;
+          // Propagate hidden -> output
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < layer1_size; c++) syn1[c + l2] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1, c + l2);
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+	    if(word_to_group != NULL && word_to_group[word] != -1){
+		target = word;
+		while(target == word) {
+			target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+            		next_random = next_random * (unsigned long long)25214903917 + 11;
+		}
+		//printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+	    }
+	    else{
+            	target = table[(next_random >> 16) % table_size];
+	    }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1neg[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha;
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1neg[c + l2] += g * neu1[c];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1neg, c + l2);
+        }
+        // Noise Contrastive Estimation
+        if (nce > 0) for (d = 0; d < nce + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+	    if(word_to_group != NULL && word_to_group[word] != -1){
+		target = word;
+		while(target == word) {
+			target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+            		next_random = next_random * (unsigned long long)25214903917 + 11;
+		}
+	    }
+	    else{
+            	target = table[(next_random >> 16) % table_size];
+	    }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+	  
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1nce[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else {
+                f = exp(f);
+                g = (label - f/(noise_distribution[target]*nce + f)) * alpha;
+          }
+	  for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1nce[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1nce[c + l2] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1nce,c + l2);
+        }
+        // hidden -> in
+        for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {
+          c = sentence_position - window + a;
+          if (c < 0) continue;
+          if (c >= sentence_length) continue;
+          last_word = sen[c];
+          if (last_word == -1) continue;
+	  BackwardCNgramWordRepresentation(neu1, vocab[last_word].word, neu1e);
+        }
+      }
+    } else if(type==1) {  //train skip-gram
+      for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        l1 = last_word * layer1_size;
+	ForwardCNgramWordRepresentation(neu1, vocab[last_word].word);
+        for (c = 0; c < layer1_size; c++) neu1e[c] = 0;
+        // HIERARCHICAL SOFTMAX
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * layer1_size;
+          // Propagate hidden -> output
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < layer1_size; c++) syn1[c + l2] += g * neu1[c];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1, c + l2);
+        }
+        // NEGATIVE SAMPLING
+	if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	    next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1neg[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha;
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1neg[c + l2] += g * neu1[c];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1neg, c + l2);
+        }
+	//Noise Contrastive Estimation
+        if (nce > 0) for (d = 0; d < nce + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	    next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1nce[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else {
+                f = exp(f);
+                g = (label - f/(noise_distribution[target]*nce + f)) * alpha;
+          }
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1nce[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1nce[c + l2] += g * neu1[c];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1nce, c + l2);
+        }
+        // Learn weights input -> hidden
+	BackwardCNgramWordRepresentation(neu1, vocab[last_word].word, neu1e);
+      }
+    }
+    else if(type == 2){ //train the cwindow architecture
+      // in -> hidden
+      cw = 0;
+      for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        window_offset = a*layer1_size;
+        if (a > window) window_offset-=layer1_size;
+	ForwardCNgramWordRepresentation(&neu1[window_offset], vocab[last_word].word);
+        cw++;
+      }
+      if (cw) {
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * window_layer_size;
+          // Propagate hidden -> output
+          for (c = 0; c < window_layer_size; c++) f += neu1[c] * syn1_window[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < window_layer_size; c++) neu1e[c] += g * syn1_window[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < window_layer_size; c++) syn1_window[c + l2] += g * neu1[c];
+	  if (cap == 1) for (c = 0; c < window_layer_size; c++) capParam(syn1_window, c + l2);
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < window_layer_size; c++) f += neu1[c] * syn1neg_window[c + l2];
+	  if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha;
+          for (c = 0; c < window_layer_size; c++) neu1e[c] += g * syn1neg_window[c + l2];
+          for (c = 0; c < window_layer_size; c++) syn1neg_window[c + l2] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < window_layer_size; c++) capParam(syn1neg_window, c + l2);
+        }
+	// Noise Contrastive Estimation
+        if (nce > 0) for (d = 0; d < nce + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < window_layer_size; c++) f += neu1[c] * syn1nce_window[c + l2];
+	  if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else {
+                f = exp(f);
+                g = (label - f/(noise_distribution[target]*nce + f)) * alpha;
+          }
+          for (c = 0; c < window_layer_size; c++) neu1e[c] += g * syn1nce_window[c + l2];
+          for (c = 0; c < window_layer_size; c++) syn1nce_window[c + l2] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < window_layer_size; c++) capParam(syn1nce_window, c + l2);
+        }
+        // hidden -> in
+        for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+          c = sentence_position - window + a;
+          if (c < 0) continue;
+          if (c >= sentence_length) continue;
+          last_word = sen[c];
+          if (last_word == -1) continue;
+	  window_offset = a * layer1_size;
+	  if(a > window) window_offset -= layer1_size;
+	  BackwardCNgramWordRepresentation(&neu1[window_offset], vocab[last_word].word, &neu1e[window_offset]);
+        }
+      }
+    }
+    else if (type == 3){  //train structured skip-gram
+      for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        l1 = last_word * layer1_size;
+	window_offset = a * layer1_size;
+	if(a > window) window_offset -= layer1_size;
+        ForwardCNgramWordRepresentation(neu1, vocab[last_word].word);
+        for (c = 0; c < layer1_size; c++) neu1e[c] = 0;
+        // HIERARCHICAL SOFTMAX
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * window_layer_size;
+          // Propagate hidden -> output
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1_window[c + l2 + window_offset];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1_window[c + l2 + window_offset];
+          // Learn weights hidden -> output
+          for (c = 0; c < layer1_size; c++) syn1[c + l2 + window_offset] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1, c + l2 + window_offset);
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	     next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1neg_window[c + l2 + window_offset];
+	  if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha;
+	  for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg_window[c + l2 + window_offset];
+          for (c = 0; c < layer1_size; c++) syn1neg_window[c + l2 + window_offset] += g * neu1[c]; 
+	  if(cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1neg_window, c + l2 + window_offset);
+        }
+	// Noise Constrastive Estimation
+        if (nce > 0) for (d = 0; d < nce + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+             next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1nce_window[c + l2 + window_offset];
+	  if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else {
+                f = exp(f);
+                g = (label - f/(noise_distribution[target]*nce + f)) * alpha;
+          }
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1nce_window[c + l2 + window_offset];
+          for (c = 0; c < layer1_size; c++) syn1nce_window[c + l2 + window_offset] += g * neu1[c];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1nce_window, c + l2 + window_offset);
+        }
+        // Learn weights input -> hidden
+        BackwardCNgramWordRepresentation(neu1, vocab[last_word].word, neu1e);
+      }
+    }
+    else if(type == 4){ //training senna
+	// in -> hidden
+      cw = 0;
+      for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        window_offset = a*layer1_size;
+        if (a > window) window_offset-=layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1[c+window_offset] += syn0[c + last_word * layer1_size];
+        cw++;
+      }
+      if (cw) {
+		for (a = 0; a < window_hidden_size; a++){
+          c = a*window_layer_size;
+          for(b = 0; b < window_layer_size; b++){
+             neu2[a] += syn_window_hidden[c + b] * neu1[b];
+          }
+        }
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * window_hidden_size;
+          // Propagate hidden -> output
+          for (c = 0; c < window_hidden_size; c++) f += hardTanh(neu2[c]) * syn_hidden_word[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < window_hidden_size; c++) neu2e[c] += dHardTanh(neu2[c],g) * g * syn_hidden_word[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < window_hidden_size; c++) syn_hidden_word[c + l2] += dHardTanh(neu2[c],g) * g * neu2[c];
+        }
+      // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	    next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_hidden_size;
+          f = 0;
+          for (c = 0; c < window_hidden_size; c++) f += hardTanh(neu2[c]) * syn_hidden_word_neg[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha / negative;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha / negative;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha / negative;
+          for (c = 0; c < window_hidden_size; c++) neu2e[c] += dHardTanh(neu2[c],g) * g * syn_hidden_word_neg[c + l2];
+          for (c = 0; c < window_hidden_size; c++) syn_hidden_word_neg[c + l2] += dHardTanh(neu2[c],g) * g * neu2[c];
+        }
+        for (a = 0; a < window_hidden_size; a++)
+          for(b = 0; b < window_layer_size; b++)
+	     neu1e[b] += neu2e[a] * syn_window_hidden[a*window_layer_size + b];
+	for (a = 0; a < window_hidden_size; a++)
+          for(b = 0; b < window_layer_size; b++)
+	     syn_window_hidden[a*window_layer_size + b] += neu2e[a] * neu1[b];
+        // hidden -> in
+        for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+          c = sentence_position - window + a;
+          if (c < 0) continue;
+          if (c >= sentence_length) continue;
+          last_word = sen[c];
+          if (last_word == -1) continue;
+          window_offset = a * layer1_size;
+          if(a > window) window_offset -= layer1_size;
+          for (c = 0; c < layer1_size; c++) syn0[c + last_word * layer1_size] += neu1e[c + window_offset];
+        }
+      }
+    }
+    else{
+	printf("unknown type %i", type);
+	exit(0);
+    }
+    sentence_position++;
+    if (sentence_position >= sentence_length) {
+      sentence_length = 0;
+      continue;
+    }
+  }
+  fclose(fi);
+  free(neu1);
+  free(neu1e);
+  pthread_exit(NULL);
+}
+
+void TrainModel() {
+  long a, b;
+  long extra_words;
+  FILE *fo;
+  FILE *fi;
+  pthread_t *pt = (pthread_t *)malloc(num_threads * sizeof(pthread_t));
+  printf("Starting training using file %s\n", train_file);
+  starting_alpha = alpha;
+  if (read_vocab_file[0] != 0) ReadVocab(); else LearnVocabFromTrainFile();
+  if (save_vocab_file[0] != 0) SaveVocab();
+  if (output_file[0] == 0) return;
+  InitNet();
+  if (negative > 0 || nce > 0) InitUnigramTable();
+  if (negative_classes_file[0] != 0) InitClassUnigramTable();
+  start = clock();
+  for (a = 0; a < num_threads; a++) pthread_create(&pt[a], NULL, TrainModelThread, (void *)a);
+  for (a = 0; a < num_threads; a++) pthread_join(pt[a], NULL);
+  fo = fopen(output_file, "wb");
+  if (classes == 0) {
+    // Save the word vectors
+    real neu1[layer1_size];
+
+    // count extra words
+    extra_words = 0;
+    fi = fopen(extra_vocab_file, "rb");
+    if (fi != NULL) {
+        char word[MAX_STRING];
+        while(1){
+		ReadWord(word, fi);
+		if(feof(fi)) break;
+		extra_words++;
+		ReadWord(word, fi);
+	}
+    }
+    fclose(fi);
+    fprintf(fo, "%lld %lld\n", vocab_size + extra_words, layer1_size);
+    for (a = 0; a < vocab_size; a++) {
+      fprintf(fo, "%s ", vocab[a].word);
+      for (b = 0; b < layer1_size; b++) neu1[b] = 0;
+      ForwardCNgramWordRepresentation(neu1, vocab[a].word);
+      
+      if (binary) for (b = 0; b < layer1_size; b++) fwrite(&neu1[b], sizeof(real), 1, fo);
+      else for (b = 0; b < layer1_size; b++) fprintf(fo, "%lf ", neu1[b]);
+      fprintf(fo, "\n");
+    }
+    fi = fopen(extra_vocab_file, "rb");
+    if (fi != NULL) {
+	char word[MAX_STRING];
+	while(1){	
+		ReadWord(word, fi);
+		if(feof(fi)) break;
+		for (b = 0; b < layer1_size; b++) neu1[b] = 0;
+		fprintf(fo, "%s ", word);
+		ForwardCNgramWordRepresentation(neu1, word);
+		if (binary) for (b = 0; b < layer1_size; b++) fwrite(&neu1[b], sizeof(real), 1, fo);
+     		else for (b = 0; b < layer1_size; b++) fprintf(fo, "%lf ", neu1[b]);
+      		fprintf(fo, "\n");
+		ReadWord(word, fi);
+	}
+    }
+    fclose(fi);
+  }
+  fclose(fo);
+}
+
+int ArgPos(char *str, int argc, char **argv) {
+  int a;
+  for (a = 1; a < argc; a++) if (!strcmp(str, argv[a])) {
+    if (a == argc - 1) {
+      printf("Argument missing for %s\n", str);
+      exit(1);
+    }
+    return a;
+  }
+  return -1;
+}
+
+int main(int argc, char **argv) {
+  int i;
+  if (argc == 1) {
+    printf("WORD VECTOR estimation toolkit v 0.1c\n\n");
+    printf("Options:\n");
+    printf("Parameters for training:\n");
+    printf("\t-train <file>\n");
+    printf("\t\tUse text data from <file> to train the model\n");
+    printf("\t-output <file>\n");
+    printf("\t\tUse <file> to save the resulting word vectors / word clusters\n");
+    printf("\t-size <int>\n");
+    printf("\t\tSet size of word vectors; default is 100\n");
+    printf("\t-window <int>\n");
+    printf("\t\tSet max skip length between words; default is 5\n");
+    printf("\t-sample <float>\n");
+    printf("\t\tSet threshold for occurrence of words. Those that appear with higher frequency in the training data\n");
+    printf("\t\twill be randomly down-sampled; default is 1e-3, useful range is (0, 1e-5)\n");
+    printf("\t-hs <int>\n");
+    printf("\t\tUse Hierarchical Softmax; default is 0 (not used)\n");
+    printf("\t-negative <int>\n");
+    printf("\t\tNumber of negative examples; default is 5, common values are 3 - 10 (0 = not used)\n");
+    printf("\t-negative-classes <file>\n");
+    printf("\t\tNegative classes to sample from\n");
+    printf("\t-nce <int>\n");
+    printf("\t\tNumber of negative examples for nce; default is 5, common values are 3 - 10 (0 = not used)\n");
+    printf("\t-threads <int>\n");
+    printf("\t\tUse <int> threads (default 12)\n");
+    printf("\t-iter <int>\n");
+    printf("\t\tRun more training iterations (default 5)\n");
+    printf("\t-cngram-size <int>\n");
+    printf("\t\tUse <int> size of the character ngrams (default 4)\n");
+    printf("\t-extra_vocab_file <file>\n");
+    printf("\t\tUse <file> file with extra words (one per line)\n");
+    printf("\t-min-count <int>\n");
+    printf("\t\tThis will discard words that appear less than <int> times; default is 5\n");
+    printf("\t-alpha <float>\n");
+    printf("\t\tSet the starting learning rate; default is 0.025 for skip-gram and 0.05 for CBOW\n");
+    printf("\t-classes <int>\n");
+    printf("\t\tOutput word classes rather than word vectors; default number of classes is 0 (vectors are written)\n");
+    printf("\t-debug <int>\n");
+    printf("\t\tSet the debug mode (default = 2 = more info during training)\n");
+    printf("\t-binary <int>\n");
+    printf("\t\tSave the resulting vectors in binary moded; default is 0 (off)\n");
+    printf("\t-save-vocab <file>\n");
+    printf("\t\tThe vocabulary will be saved to <file>\n");
+    printf("\t-read-vocab <file>\n");
+    printf("\t\tThe vocabulary will be read from <file>, not constructed from the training data\n");
+    printf("\t-type <int>\n");
+    printf("\t\tType of embeddings (0 for cbow, 1 for skipngram, 2 for cwindow, 3 for structured skipngram, 4 for senna type)\n");
+    printf("\t-cap <int>\n");
+    printf("\t\tlimit the parameter values to the range [-50, 50]; default is 0 (off)\n");
+    printf("\nExamples:\n");
+    printf("./word2vec -train data.txt -output vec.txt -size 200 -window 5 -sample 1e-4 -negative 5 -hs 0 -binary 0 -type 1 -iter 3 -cngram-size 4 -extra_vocab_file extra.txt \n\n");
+    return 0;
+  }
+  output_file[0] = 0;
+  save_vocab_file[0] = 0;
+  read_vocab_file[0] = 0;
+  negative_classes_file[0] = 0;
+  if ((i = ArgPos((char *)"-size", argc, argv)) > 0) layer1_size = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-train", argc, argv)) > 0) strcpy(train_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-save-vocab", argc, argv)) > 0) strcpy(save_vocab_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-read-vocab", argc, argv)) > 0) strcpy(read_vocab_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-debug", argc, argv)) > 0) debug_mode = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-binary", argc, argv)) > 0) binary = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-type", argc, argv)) > 0) type = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-output", argc, argv)) > 0) strcpy(output_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-window", argc, argv)) > 0) window = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-sample", argc, argv)) > 0) sample = atof(argv[i + 1]);
+  if ((i = ArgPos((char *)"-hs", argc, argv)) > 0) hs = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-negative", argc, argv)) > 0) negative = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-negative-classes", argc, argv)) > 0) strcpy(negative_classes_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-nce", argc, argv)) > 0) nce = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-threads", argc, argv)) > 0) num_threads = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-iter", argc, argv)) > 0) iter = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-min-count", argc, argv)) > 0) min_count = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-classes", argc, argv)) > 0) classes = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-cap", argc, argv)) > 0) cap = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-cngram-size", argc, argv)) > 0) cngram_size = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-extra_vocab_file", argc, argv)) > 0) strcpy(extra_vocab_file, argv[i + 1]); 
+  if (type==0 || type==2 || type==4) alpha = 0.05;
+  if ((i = ArgPos((char *)"-alpha", argc, argv)) > 0) alpha = atof(argv[i + 1]);
+  vocab = (struct vocab_word *)calloc(vocab_max_size, sizeof(struct vocab_word));
+  vocab_hash = (int *)calloc(vocab_hash_size, sizeof(int));
+  cngram_vocab = (struct vocab_word *)calloc(cngram_vocab_max_size, sizeof(struct vocab_word));
+  cngram_vocab_hash = (int *)calloc(vocab_hash_size, sizeof(int));
+  expTable = (real *)malloc((EXP_TABLE_SIZE + 1) * sizeof(real));
+  for (i = 0; i < EXP_TABLE_SIZE; i++) {
+    expTable[i] = exp((i / (real)EXP_TABLE_SIZE * 2 - 1) * MAX_EXP); // Precompute the exp() table
+    expTable[i] = expTable[i] / (expTable[i] + 1);                   // Precompute f(x) = x / (x + 1)
+  }
+  TrainModel();
+  return 0;
+}
+
diff --git a/compute-accuracy.c b/compute-accuracy.c
new file mode 100644
index 0000000..95a83e4
--- /dev/null
+++ b/compute-accuracy.c
@@ -0,0 +1,143 @@
+//  Copyright 2013 Google Inc. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <stdlib.h>
+#include <ctype.h>
+
+const long long max_size = 2000;         // max length of strings
+const long long N = 1;                   // number of closest words
+const long long max_w = 50;              // max length of vocabulary entries
+
+int main(int argc, char **argv)
+{
+  FILE *f;
+  char st1[max_size], st2[max_size], st3[max_size], st4[max_size], bestw[N][max_size], file_name[max_size];
+  float dist, len, bestd[N], vec[max_size];
+  long long words, size, a, b, c, d, b1, b2, b3, threshold = 0;
+  float *M;
+  char *vocab;
+  int TCN, CCN = 0, TACN = 0, CACN = 0, SECN = 0, SYCN = 0, SEAC = 0, SYAC = 0, QID = 0, TQ = 0, TQS = 0;
+  if (argc < 2) {
+    printf("Usage: ./compute-accuracy <FILE> <threshold>\nwhere FILE contains word projections, and threshold is used to reduce vocabulary of the model for fast approximate evaluation (0 = off, otherwise typical value is 30000)\n");
+    return 0;
+  }
+  strcpy(file_name, argv[1]);
+  if (argc > 2) threshold = atoi(argv[2]);
+  f = fopen(file_name, "rb");
+  if (f == NULL) {
+    printf("Input file not found\n");
+    return -1;
+  }
+  fscanf(f, "%lld", &words);
+  if (threshold) if (words > threshold) words = threshold;
+  fscanf(f, "%lld", &size);
+  vocab = (char *)malloc(words * max_w * sizeof(char));
+  M = (float *)malloc(words * size * sizeof(float));
+  if (M == NULL) {
+    printf("Cannot allocate memory: %lld MB\n", words * size * sizeof(float) / 1048576);
+    return -1;
+  }
+  for (b = 0; b < words; b++) {
+    a = 0;
+    while (1) {
+      vocab[b * max_w + a] = fgetc(f);
+      if (feof(f) || (vocab[b * max_w + a] == ' ')) break;
+      if ((a < max_w) && (vocab[b * max_w + a] != '\n')) a++;
+    }
+    vocab[b * max_w + a] = 0;
+    for (a = 0; a < max_w; a++) vocab[b * max_w + a] = toupper(vocab[b * max_w + a]);
+    for (a = 0; a < size; a++) fread(&M[a + b * size], sizeof(float), 1, f);
+    len = 0;
+    for (a = 0; a < size; a++) len += M[a + b * size] * M[a + b * size];
+    len = sqrt(len);
+    for (a = 0; a < size; a++) M[a + b * size] /= len;
+  }
+  fclose(f);
+  TCN = 0;
+  while (1) {
+    for (a = 0; a < N; a++) bestd[a] = 0;
+    for (a = 0; a < N; a++) bestw[a][0] = 0;
+    scanf("%s", st1);
+    for (a = 0; a < strlen(st1); a++) st1[a] = toupper(st1[a]);
+    if ((!strcmp(st1, ":")) || (!strcmp(st1, "EXIT")) || feof(stdin)) {
+      if (TCN == 0) TCN = 1;
+      if (QID != 0) {
+        printf("ACCURACY TOP1: %.2f %%  (%d / %d)\n", CCN / (float)TCN * 100, CCN, TCN);
+        printf("Total accuracy: %.2f %%   Semantic accuracy: %.2f %%   Syntactic accuracy: %.2f %% \n", CACN / (float)TACN * 100, SEAC / (float)SECN * 100, SYAC / (float)SYCN * 100);
+      }
+      QID++;
+      scanf("%s", st1);
+      if (feof(stdin)) break;
+      printf("%s:\n", st1);
+      TCN = 0;
+      CCN = 0;
+      continue;
+    }
+    if (!strcmp(st1, "EXIT")) break;
+    scanf("%s", st2);
+    for (a = 0; a < strlen(st2); a++) st2[a] = toupper(st2[a]);
+    scanf("%s", st3);
+    for (a = 0; a<strlen(st3); a++) st3[a] = toupper(st3[a]);
+    scanf("%s", st4);
+    for (a = 0; a < strlen(st4); a++) st4[a] = toupper(st4[a]);
+    for (b = 0; b < words; b++) if (!strcmp(&vocab[b * max_w], st1)) break;
+    b1 = b;
+    for (b = 0; b < words; b++) if (!strcmp(&vocab[b * max_w], st2)) break;
+    b2 = b;
+    for (b = 0; b < words; b++) if (!strcmp(&vocab[b * max_w], st3)) break;
+    b3 = b;
+    for (a = 0; a < N; a++) bestd[a] = 0;
+    for (a = 0; a < N; a++) bestw[a][0] = 0;
+    TQ++;
+    if (b1 == words) continue;
+    if (b2 == words) continue;
+    if (b3 == words) continue;
+    for (b = 0; b < words; b++) if (!strcmp(&vocab[b * max_w], st4)) break;
+    if (b == words) continue;
+    for (a = 0; a < size; a++) vec[a] = (M[a + b2 * size] - M[a + b1 * size]) + M[a + b3 * size];
+    TQS++;
+    for (c = 0; c < words; c++) {
+      if (c == b1) continue;
+      if (c == b2) continue;
+      if (c == b3) continue;
+      dist = 0;
+      for (a = 0; a < size; a++) dist += vec[a] * M[a + c * size];
+      for (a = 0; a < N; a++) {
+        if (dist > bestd[a]) {
+          for (d = N - 1; d > a; d--) {
+            bestd[d] = bestd[d - 1];
+            strcpy(bestw[d], bestw[d - 1]);
+          }
+          bestd[a] = dist;
+          strcpy(bestw[a], &vocab[c * max_w]);
+          break;
+        }
+      }
+    }
+    if (!strcmp(st4, bestw[0])) {
+      CCN++;
+      CACN++;
+      if (QID <= 5) SEAC++; else SYAC++;
+    }
+    if (QID <= 5) SECN++; else SYCN++;
+    TCN++;
+    TACN++;
+  }
+  printf("Questions seen / total: %d %d   %.2f %% \n", TQS, TQ, TQS/(float)TQ*100);
+  return 0;
+}
diff --git a/distance.c b/distance.c
new file mode 100644
index 0000000..e01ba9b
--- /dev/null
+++ b/distance.c
@@ -0,0 +1,164 @@
+//  Copyright 2013 Google Inc. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+
+#include <stdio.h>
+#include <string.h>
+#include <math.h>
+#include <stdlib.h>
+
+const long long max_size = 2000;         // max length of strings
+const long long N = 40;                  // number of closest words that will be shown
+const long long max_w = 50;              // max length of vocabulary entries
+
+#define MAX_STRING 100
+void ReadWord(char *word, FILE *fin) {
+  int a = 0, ch;
+  while (!feof(fin)) {
+    ch = fgetc(fin);
+    if (ch == 13) continue;
+    if ((ch == ' ') || (ch == '\t') || (ch == '\n')) {
+      if (a > 0) {
+        if (ch == '\n') ungetc(ch, fin);
+        break;
+      }
+      if (ch == '\n') {
+        strcpy(word, (char *)"</s>");
+        return;
+      } else continue;
+    }
+    word[a] = ch;
+    a++;
+    if (a >= MAX_STRING - 1) a--;   // Truncate too long words
+  }
+  word[a] = 0;
+}
+
+int main(int argc, char **argv) {
+  FILE *f;
+  char st1[max_size];
+  char *bestw[N];
+  char file_name[max_size], st[100][max_size];
+  float dist, len, bestd[N], vec[max_size];
+  long long words, size, a, b, c, d, cn, bi[100];
+  float *M;
+  char *vocab;
+  if (argc < 2) {
+    printf("Usage: ./distance <FILE>\nwhere FILE contains word projections in the BINARY FORMAT\n");
+    return 0;
+  }
+  strcpy(file_name, argv[1]);
+  f = fopen(file_name, "rb");
+  if (f == NULL) {
+    printf("Input file not found\n");
+    return -1;
+  }
+  fscanf(f, "%lld", &words);
+  fscanf(f, "%lld", &size);
+  vocab = (char *)malloc((long long)words * max_w * sizeof(char));
+  for (a = 0; a < N; a++) bestw[a] = (char *)malloc(max_size * sizeof(char));
+  M = (float *)malloc((long long)words * (long long)size * sizeof(float));
+  if (M == NULL) {
+    printf("Cannot allocate memory: %lld MB    %lld  %lld\n", (long long)words * size * sizeof(float) / 1048576, words, size);
+    return -1;
+  }
+  for (b = 0; b < words; b++) {
+    a = 0;
+    while (1) {
+      vocab[b * max_w + a] = fgetc(f);
+      if (feof(f) || (vocab[b * max_w + a] == ' ')) break;
+      if ((a < max_w) && (vocab[b * max_w + a] != '\n')) a++;
+    }
+    vocab[b * max_w + a] = 0;
+    for (a = 0; a < size; a++) fread(&M[a + b * size], sizeof(float), 1, f);
+    len = 0;
+    for (a = 0; a < size; a++) len += M[a + b * size] * M[a + b * size];
+    len = sqrt(len);
+    for (a = 0; a < size; a++) M[a + b * size] /= len;
+  }
+  fclose(f);
+  while (1) {
+    for (a = 0; a < N; a++) bestd[a] = 0;
+    for (a = 0; a < N; a++) bestw[a][0] = 0;
+    printf("Enter word or sentence (EXIT to break): ");
+    a = 0;
+    while (1) {
+      st1[a] = fgetc(stdin);
+      if ((st1[a] == '\n') || (a >= max_size - 1)) {
+        st1[a] = 0;
+        break;
+      }
+      a++;
+    }
+    if (!strcmp(st1, "EXIT")) break;
+    cn = 0;
+    b = 0;
+    c = 0;
+    while (1) {
+      st[cn][b] = st1[c];
+      b++;
+      c++;
+      st[cn][b] = 0;
+      if (st1[c] == 0) break;
+      if (st1[c] == ' ') {
+        cn++;
+        b = 0;
+        c++;
+      }
+    }
+    cn++;
+    for (a = 0; a < cn; a++) {
+      for (b = 0; b < words; b++) if (!strcmp(&vocab[b * max_w], st[a])) break;
+      if (b == words) b = -1;
+      bi[a] = b;
+      printf("\nWord: %s  Position in vocabulary: %lld\n", st[a], bi[a]);
+      if (b == -1) {
+        printf("Out of dictionary word!\n");
+        break;
+      }
+    }
+    if (b == -1) continue;
+    printf("\n                                              Word       Cosine distance\n------------------------------------------------------------------------\n");
+    for (a = 0; a < size; a++) vec[a] = 0;
+    for (b = 0; b < cn; b++) {
+      if (bi[b] == -1) continue;
+      for (a = 0; a < size; a++) vec[a] += M[a + bi[b] * size];
+    }
+    len = 0;
+    for (a = 0; a < size; a++) len += vec[a] * vec[a];
+    len = sqrt(len);
+    for (a = 0; a < size; a++) vec[a] /= len;
+    for (a = 0; a < N; a++) bestd[a] = -1;
+    for (a = 0; a < N; a++) bestw[a][0] = 0;
+    for (c = 0; c < words; c++) {
+      a = 0;
+      for (b = 0; b < cn; b++) if (bi[b] == c) a = 1;
+      if (a == 1) continue;
+      dist = 0;
+      for (a = 0; a < size; a++) dist += vec[a] * M[a + c * size];
+      for (a = 0; a < N; a++) {
+        if (dist > bestd[a]) {
+          for (d = N - 1; d > a; d--) {
+            bestd[d] = bestd[d - 1];
+            strcpy(bestw[d], bestw[d - 1]);
+          }
+          bestd[a] = dist;
+          strcpy(bestw[a], &vocab[c * max_w]);
+          break;
+        }
+      }
+    }
+    for (a = 0; a < N; a++) printf("%50s\t\t%f\n", bestw[a], bestd[a]);
+  }
+  return 0;
+}
diff --git a/distance_fast.c b/distance_fast.c
new file mode 100644
index 0000000..8b143e9
--- /dev/null
+++ b/distance_fast.c
@@ -0,0 +1,261 @@
+//  Copyright 2013 Google Inc. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+
+#include <stdio.h>
+#include <string.h>
+#include <math.h>
+#include <stdlib.h>
+#include <time.h>
+
+const long long max_size = 2000;         // max length of strings
+const long long N = 10;                  // number of closest words that will be shown
+const long long max_w = 50;              // max length of vocabulary entries
+
+#define MAX_STRING 100
+void ReadWord(char *word, FILE *fin) {
+  int a = 0, ch;
+  while (!feof(fin)) {
+    ch = fgetc(fin);
+    if (ch == 13) continue;
+    if ((ch == ' ') || (ch == '\t') || (ch == '\n')) {
+      if (a > 0) {
+        if (ch == '\n') ungetc(ch, fin);
+        break;
+      }
+      if (ch == '\n') {
+        strcpy(word, (char *)"</s>");
+        return;
+      } else continue;
+    }
+    word[a] = ch;
+    a++;
+    if (a >= MAX_STRING - 1) a--;   // Truncate too long words
+  }
+  word[a] = 0;
+}
+
+int main(int argc, char **argv) {
+  FILE *f;
+  char st1[max_size];
+  char *bestw[N];
+  char file_name[max_size], st[100][max_size];
+  float dist, len, bestd[N], bestclasses[N], vec[max_size];
+  int bestclasses_ids[N];
+  long long words, size, a, b, c, d, e, cn, bi[100];
+  float *M;
+  char *vocab;
+  char word[MAX_STRING];
+  clock_t begin;
+  if (argc < 2) {
+    printf("Usage: ./kmeans_txt <FILE>\nwhere FILE contains features\n <number_of_classes>");
+    return 0;
+  }
+  strcpy(file_name, argv[1]);
+  int classes = atoi(argv[2]);
+  f = fopen(file_name, "rb");
+  if (f == NULL) {
+    printf("Input file not found\n");
+    return -1;
+  }
+  
+  printf("reading data\n");
+  ReadWord(word, f);
+  words = atoi(word);
+  ReadWord(word, f);
+  size = atoi(word);
+  vocab = (char *)malloc((long long)words * max_w * sizeof(char));
+  for (a = 0; a < N; a++) bestw[a] = (char *)malloc(max_size * sizeof(char));
+  M = (float *)malloc((long long)words * (long long)size * sizeof(float));
+  if (M == NULL) {
+    printf("Cannot allocate memory: %lld MB    %lld  %lld\n", (long long)words * size * sizeof(float) / 1048576, words, size);
+    return -1;
+  }
+  for (b = 0; b < words; b++) {
+    a = 0;
+    while (1) {
+      vocab[b * max_w + a] = fgetc(f);
+      if (feof(f) || (vocab[b * max_w + a] == ' ')) break;
+      if ((a < max_w) && (vocab[b * max_w + a] != '\n')) a++;
+    }
+    vocab[b * max_w + a] = 0;
+    for (a = 0; a < size; a++) {
+        ReadWord(word,f); 
+        M[a + b * size] = atof(word); 
+    }
+    len = 0;
+    for (a = 0; a < size; a++) len += M[a + b * size] * M[a + b * size];
+    len = sqrt(len);
+    for (a = 0; a < size; a++) M[a + b * size] /= len;
+  }
+  fclose(f);
+  
+  //run kmeans
+  printf("running k-means with %i classes...\n",classes);
+  int clcn = classes, iter = 10, closeid;
+  int *centcn = (int *)malloc(classes * sizeof(int));
+  int *cl = (int *)calloc(words, sizeof(int));
+  float closev, x;
+  float *cent = (float *)calloc(classes * size, sizeof(float));
+  for (a = 0; a < words; a++) cl[a] = a % clcn;
+  for (a = 0; a < iter; a++) {
+    for (b = 0; b < clcn * size; b++) cent[b] = 0;
+    for (b = 0; b < clcn; b++) centcn[b] = 1;
+    for (c = 0; c < words; c++) {
+      for (d = 0; d < size; d++) cent[size * cl[c] + d] += M[c * size + d];
+      centcn[cl[c]]++;
+    }
+    for (b = 0; b < clcn; b++) {
+      closev = 0;
+      for (c = 0; c < size; c++) {
+        cent[size * b + c] /= centcn[b];
+        closev += cent[size * b + c] * cent[size * b + c];
+      }
+      closev = sqrt(closev);
+      for (c = 0; c < size; c++) cent[size * b + c] /= closev;
+    }
+    for (c = 0; c < words; c++) {
+      closev = -10;
+      closeid = 0;
+      for (d = 0; d < clcn; d++) {
+        x = 0;
+        for (b = 0; b < size; b++) x += cent[size * d + b] * M[c * size + b];
+        if (x > closev) {
+          closev = x;
+          closeid = d;
+        }
+      }
+      cl[c] = closeid;
+    }
+  }
+  
+  // build an array of words ordered by class and their offsets (index where each class starts)
+  int class_words[words];
+  int class_offsets[classes];
+  for(a = 0; a < classes; a++) class_offsets[a]=0;
+  for(a = 0; a < words; a++) class_offsets[cl[a]]++;
+  for(a = 1; a < classes; a++) class_offsets[a] += class_offsets[a-1];
+  for(a = 0; a < words; a++) class_words[--class_offsets[cl[a]]] = a;   
+  
+  //reading from input
+  while (1) {
+    for (a = 0; a < N; a++) bestd[a] = 0;
+    for (a = 0; a < N; a++) bestclasses[a] = 0;
+    for (a = 0; a < N; a++) bestw[a][0] = 0;
+    printf("Enter word or sentence (EXIT to break): ");
+    a = 0;
+    while (1) {
+      st1[a] = fgetc(stdin);
+      if ((st1[a] == '\n') || (a >= max_size - 1)) {
+        st1[a] = 0;
+        break;
+      }
+      a++;
+    }
+    if (!strcmp(st1, "EXIT")) break;
+    cn = 0;
+    b = 0;
+    c = 0;
+    while (1) {
+      st[cn][b] = st1[c];
+      b++;
+      c++;
+      st[cn][b] = 0;
+      if (st1[c] == 0) break;
+      if (st1[c] == ' ') {
+        cn++;
+        b = 0;
+        c++;
+      }
+    }
+    cn++;
+    for (a = 0; a < cn; a++) {
+      for (b = 0; b < words; b++) if (!strcmp(&vocab[b * max_w], st[a])) break;
+      if (b == words) b = -1;
+      bi[a] = b;
+      printf("\nWord: %s  Position in vocabulary: %lld\n", st[a], bi[a]);
+      if (b == -1) {
+        printf("Out of dictionary word!\n");
+        break;
+      }
+    }
+    if (b == -1) continue;
+    begin = clock();
+    
+    printf("\n                                              Word       Cosine distance\n------------------------------------------------------------------------\n");
+    
+    for (a = 0; a < size; a++) vec[a] = 0;
+    for (b = 0; b < cn; b++) {
+      if (bi[b] == -1) continue;
+      for (a = 0; a < size; a++) vec[a] += M[a + bi[b] * size];
+    }
+    
+    len = 0;
+    for (a = 0; a < size; a++) len += vec[a] * vec[a];
+    len = sqrt(len);
+    for (a = 0; a < size; a++) vec[a] /= len;
+    
+    // find top N centroids
+    for (a = 0; a < N; a++) bestclasses[a] = -1;
+    for (a = 0; a < N; a++) bestclasses_ids[a] = -1;
+    for (c = 0; c < classes; c++){
+    	dist = 0;
+    	for (a = 0; a < size; a++) dist += vec[a] * cent[a + size * c];
+        for (a = 0; a < N; a++) {
+        if (dist > bestclasses[a]) {
+          	for(d = N - 1; d > a; d--){
+          		bestclasses[d] = bestclasses[d-1];
+          		bestclasses_ids[d] = bestclasses_ids[d-1];
+          	}
+          	bestclasses[a] = dist;
+          	bestclasses_ids[a] = c;
+          	break;
+        }
+    	}
+    }
+    
+    // find top N words in the centroids
+    for (a = 0; a < N; a++) bestd[a] = -1;
+    for (a = 0; a < N; a++) bestw[a][0] = 0;
+    for (a = 0; a < N; a++){
+   		c = words;
+   		if(bestclasses_ids[a] < classes-1) c = class_offsets[bestclasses_ids[a]+1];
+   		b = class_offsets[bestclasses_ids[a]];
+   		for(; b < c; b++){
+   			dist = 0;
+            for (d = 0; d < size; d++) dist += vec[d] * M[d + class_words[b] * size];
+            for (d = 0; d < N; d++){
+            	if(dist > bestd[d]){
+            		for (e = N -1; e > d; e--){
+            			bestd[e] = bestd[e-1];
+            			strcpy(bestw[e], bestw[e-1]);
+            		}
+            		bestd[d] = dist;
+            		strcpy(bestw[d], &vocab[class_words[b] * max_w]);
+            		break;
+            	}
+            }
+   	    }
+   	}
+	for (a = 0; a < N; a++) printf("%50s\t\t%f\n", bestw[a], bestd[a]);   
+	printf("time spent = %f seconds\n", (double)(clock() - begin) / CLOCKS_PER_SEC);
+  }
+  // Save the K-means classes
+
+  free(centcn);
+  free(cent);
+  free(cl);
+  
+  //start running distance
+  return 0;
+}
diff --git a/distance_txt.c b/distance_txt.c
new file mode 100644
index 0000000..da5b67e
--- /dev/null
+++ b/distance_txt.c
@@ -0,0 +1,175 @@
+//  Copyright 2013 Google Inc. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+
+#include <stdio.h>
+#include <string.h>
+#include <math.h>
+#include <stdlib.h>
+#include <time.h>
+
+const long long max_size = 2000;         // max length of strings
+const long long N = 40;                  // number of closest words that will be shown
+const long long max_w = 50;              // max length of vocabulary entries
+
+#define MAX_STRING 100
+void ReadWord(char *word, FILE *fin) {
+  int a = 0, ch;
+  while (!feof(fin)) {
+    ch = fgetc(fin);
+    if (ch == 13) continue;
+    if ((ch == ' ') || (ch == '\t') || (ch == '\n')) {
+      if (a > 0) {
+        if (ch == '\n') ungetc(ch, fin);
+        break;
+      }
+      if (ch == '\n') {
+        strcpy(word, (char *)"</s>");
+        return;
+      } else continue;
+    }
+    word[a] = ch;
+    a++;
+    if (a >= MAX_STRING - 1) a--;   // Truncate too long words
+  }
+  word[a] = 0;
+}
+
+int main(int argc, char **argv) {
+  FILE *f;
+  char st1[max_size];
+  char *bestw[N];
+  char file_name[max_size], st[100][max_size];
+  float dist, len, bestd[N], vec[max_size];
+  long long words, size, a, b, c, d, cn, bi[100];
+  float *M;
+  char *vocab;
+  char word[MAX_STRING];
+  clock_t begin;
+  if (argc < 2) {
+    printf("Usage: ./distance <FILE>\nwhere FILE contains word projections in the BINARY FORMAT\n");
+    return 0;
+  }
+  strcpy(file_name, argv[1]);
+  f = fopen(file_name, "rb");
+  if (f == NULL) {
+    printf("Input file not found\n");
+    return -1;
+  }
+  ReadWord(word, f);
+  words = atoi(word);
+  ReadWord(word, f);
+  size = atoi(word);
+  vocab = (char *)malloc((long long)words * max_w * sizeof(char));
+  for (a = 0; a < N; a++) bestw[a] = (char *)malloc(max_size * sizeof(char));
+  M = (float *)malloc((long long)words * (long long)size * sizeof(float));
+  if (M == NULL) {
+    printf("Cannot allocate memory: %lld MB    %lld  %lld\n", (long long)words * size * sizeof(float) / 1048576, words, size);
+    return -1;
+  }
+  for (b = 0; b < words; b++) {
+    a = 0;
+    while (1) {
+      vocab[b * max_w + a] = fgetc(f);
+      if (feof(f) || (vocab[b * max_w + a] == ' ')) break;
+      if ((a < max_w) && (vocab[b * max_w + a] != '\n')) a++;
+    }
+    vocab[b * max_w + a] = 0;
+    for (a = 0; a < size; a++) {
+        ReadWord(word,f); 
+        M[a + b * size] = atof(word); 
+    }
+    len = 0;
+    for (a = 0; a < size; a++) len += M[a + b * size] * M[a + b * size];
+    len = sqrt(len);
+    for (a = 0; a < size; a++) M[a + b * size] /= len;
+  }
+  fclose(f);
+  while (1) {
+    for (a = 0; a < N; a++) bestd[a] = 0;
+    for (a = 0; a < N; a++) bestw[a][0] = 0;
+    printf("Enter word or sentence (EXIT to break): ");
+    a = 0;
+    while (1) {
+      st1[a] = fgetc(stdin);
+      if ((st1[a] == '\n') || (a >= max_size - 1)) {
+        st1[a] = 0;
+        break;
+      }
+      a++;
+    }
+    if (!strcmp(st1, "EXIT")) break;
+    cn = 0;
+    b = 0;
+    c = 0;
+    while (1) {
+      st[cn][b] = st1[c];
+      b++;
+      c++;
+      st[cn][b] = 0;
+      if (st1[c] == 0) break;
+      if (st1[c] == ' ') {
+        cn++;
+        b = 0;
+        c++;
+      }
+    }
+    cn++;
+    for (a = 0; a < cn; a++) {
+      for (b = 0; b < words; b++) if (!strcmp(&vocab[b * max_w], st[a])) break;
+      if (b == words) b = -1;
+      bi[a] = b;
+      printf("\nWord: %s  Position in vocabulary: %lld\n", st[a], bi[a]);
+      if (b == -1) {
+        printf("Out of dictionary word!\n");
+        break;
+      }
+    }
+    if (b == -1) continue;
+    begin = clock();
+
+    printf("\n                                              Word       Cosine distance\n------------------------------------------------------------------------\n");
+    for (a = 0; a < size; a++) vec[a] = 0;
+    for (b = 0; b < cn; b++) {
+      if (bi[b] == -1) continue;
+      for (a = 0; a < size; a++) vec[a] += M[a + bi[b] * size];
+    }
+    len = 0;
+    for (a = 0; a < size; a++) len += vec[a] * vec[a];
+    len = sqrt(len);
+    for (a = 0; a < size; a++) vec[a] /= len;
+    for (a = 0; a < N; a++) bestd[a] = -1;
+    for (a = 0; a < N; a++) bestw[a][0] = 0;
+    for (c = 0; c < words; c++) {
+      a = 0;
+      for (b = 0; b < cn; b++) if (bi[b] == c) a = 1;
+      if (a == 1) continue;
+      dist = 0;
+      for (a = 0; a < size; a++) dist += vec[a] * M[a + c * size];
+      for (a = 0; a < N; a++) {
+        if (dist > bestd[a]) {
+          for (d = N - 1; d > a; d--) {
+            bestd[d] = bestd[d - 1];
+            strcpy(bestw[d], bestw[d - 1]);
+          }
+          bestd[a] = dist;
+          strcpy(bestw[a], &vocab[c * max_w]);
+          break;
+        }
+      }
+    }
+    for (a = 0; a < N; a++) printf("%50s\t\t%f\n", bestw[a], bestd[a]);
+    printf("time spent = %f seconds\n", (double)(clock() - begin) / CLOCKS_PER_SEC);
+  }
+  return 0;
+}
diff --git a/kmeans_txt.c b/kmeans_txt.c
new file mode 100644
index 0000000..16934ab
--- /dev/null
+++ b/kmeans_txt.c
@@ -0,0 +1,161 @@
+//  Copyright 2013 Google Inc. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+
+#include <stdio.h>
+#include <string.h>
+#include <math.h>
+#include <stdlib.h>
+
+const long long max_size = 2000;         // max length of strings
+const long long N = 40;                  // number of closest words that will be shown
+const long long max_w = 50;              // max length of vocabulary entries
+
+#define MAX_STRING 100
+void ReadWord(char *word, FILE *fin) {
+  int a = 0, ch;
+  while (!feof(fin)) {
+    ch = fgetc(fin);
+    if (ch == 13) continue;
+    if ((ch == ' ') || (ch == '\t') || (ch == '\n')) {
+      if (a > 0) {
+        if (ch == '\n') ungetc(ch, fin);
+        break;
+      }
+      if (ch == '\n') {
+        strcpy(word, (char *)"</s>");
+        return;
+      } else continue;
+    }
+    word[a] = ch;
+    a++;
+    if (a >= MAX_STRING - 1) a--;   // Truncate too long words
+  }
+  word[a] = 0;
+}
+
+int main(int argc, char **argv) {
+  FILE *f;
+  char file_name[max_size], output_file[max_size];
+  float len;
+  long long words, size, a, b, c, d;
+  float *M;
+  char *vocab;
+  char word[MAX_STRING];
+  if (argc < 3) {
+    printf("Usage: ./kmeans_txt <FILE>\nwhere FILE contains features\n <number_of_classes>");
+    return 0;
+  }
+  strcpy(file_name, argv[1]);
+  strcpy(output_file, argv[2]);
+  int classes = atoi(argv[3]);
+  f = fopen(file_name, "rb");
+  if (f == NULL) {
+    printf("Input file not found\n");
+    return -1;
+  }
+  
+  FILE *fo = fopen(output_file, "wb");
+  
+  ReadWord(word, f);
+  words = atoi(word);
+  ReadWord(word, f);
+  size = atoi(word);
+  vocab = (char *)malloc((long long)words * max_w * sizeof(char));
+  M = (float *)malloc((long long)words * (long long)size * sizeof(float));
+  if (M == NULL) {
+    printf("Cannot allocate memory: %lld MB    %lld  %lld\n", (long long)words * size * sizeof(float) / 1048576, words, size);
+    return -1;
+  }
+  for (b = 0; b < words; b++) {
+    a = 0;
+    while (1) {
+      vocab[b * max_w + a] = fgetc(f);
+      if (feof(f) || (vocab[b * max_w + a] == ' ')) break;
+      if ((a < max_w) && (vocab[b * max_w + a] != '\n')) a++;
+    }
+    vocab[b * max_w + a] = 0;
+    for (a = 0; a < size; a++) {
+        ReadWord(word,f); 
+        M[a + b * size] = atof(word); 
+    }
+    len = 0;
+    for (a = 0; a < size; a++) len += M[a + b * size] * M[a + b * size];
+    len = sqrt(len);
+    for (a = 0; a < size; a++) M[a + b * size] /= len;
+  }
+  fclose(f);
+  
+  //run kmeans
+  int clcn = classes, iter = 2, closeid;
+  int *centcn = (int *)malloc(classes * sizeof(int));
+  int *cl = (int *)calloc(words, sizeof(int));
+  float closev, x;
+  float *cent = (float *)calloc(classes * size, sizeof(float));
+  for (a = 0; a < words; a++) cl[a] = a % clcn;
+    for (a = 0; a < iter; a++) {
+      for (b = 0; b < clcn * size; b++) cent[b] = 0;
+      for (b = 0; b < clcn; b++) centcn[b] = 1;
+      for (c = 0; c < words; c++) {
+        for (d = 0; d < size; d++) cent[size * cl[c] + d] += M[c * size + d];
+        centcn[cl[c]]++;
+      }
+      for (b = 0; b < clcn; b++) {
+        closev = 0;
+        for (c = 0; c < size; c++) {
+          cent[size * b + c] /= centcn[b];
+          closev += cent[size * b + c] * cent[size * b + c];
+        }
+        closev = sqrt(closev);
+        for (c = 0; c < size; c++) cent[size * b + c] /= closev;
+      }
+      for (c = 0; c < words; c++) {
+        closev = -10;
+        closeid = 0;
+        for (d = 0; d < clcn; d++) {
+          x = 0;
+          for (b = 0; b < size; b++) x += cent[size * d + b] * M[c * size + b];
+          if (x > closev) {
+            closev = x;
+            closeid = d;
+          }
+        }
+        cl[c] = closeid;
+      }
+    }
+    
+    // build an array of words ordered by class and their offsets (index where each class starts)
+    int class_words[words];
+    int class_offsets[classes];
+    for(a = 0; a < classes; a++) class_offsets[a]=0;
+    for(a = 0; a < words; a++) class_offsets[cl[a]]++;
+    for(a = 1; a < classes; a++) class_offsets[a] += class_offsets[a-1];
+    for(a = 0; a < words; a++) class_words[--class_offsets[cl[a]]] = a;
+    
+    for (a = 0; a < classes; a++){
+    	c = words;
+    	if(a < classes-1) c = class_offsets[a+1];
+    	b = class_offsets[a];
+    	for(; b < c; b++){
+    	    fprintf(fo, "%lld %s\n", a ,&vocab[class_words[b] * max_w]);
+    	}
+    }
+     // Save the K-means classes
+    //for (a = 0; a < words; a++) fprintf(fo, "%s %d\n", &vocab[a * max_w], cl[a]);
+    free(centcn);
+    free(cent);
+    free(cl);
+    free(M);
+    free(vocab);
+  return 0;
+}
diff --git a/makefile b/makefile
new file mode 100644
index 0000000..0ea6193
--- /dev/null
+++ b/makefile
@@ -0,0 +1,32 @@
+CC = gcc
+#Using -Ofast instead of -O3 might result in faster code, but is supported only by newer GCC versions
+CFLAGS = -lm -pthread -O3 -march=k8 -mtune=k8 -Wall -funroll-loops
+#CFLAGS = -m64 -march=k8 -mtune=k8 -lm -pthread -O3 -Wall -funroll-loops 
+
+
+all: word2vec cngram2vec weightedWord2vec wordless2vec word2phrase distance word-analogy compute-accuracy distance_txt distance_fast kmeans_txt
+
+word2vec : word2vecExt.c
+	$(CC) word2vecExt.c -o word2vec $(CFLAGS)
+weightedWord2vec : weightedWord2vec.c
+	$(CC) weightedWord2vec.c -o weightedWord2vec $(CFLAGS)
+cngram2vec : cngram2vec.c
+	$(CC) cngram2vec.c -o cngram2vec $(CFLAGS)
+wordless2vec : wordless2vec.c
+	$(CC) wordless2vec.c -o wordless2vec $(CFLAGS)
+word2phrase : word2phrase.c
+	$(CC) word2phrase.c -o word2phrase $(CFLAGS)
+distance : distance.c
+	$(CC) distance.c -o distance $(CFLAGS)
+distance_txt : distance_txt.c
+	$(CC) distance_txt.c -o distance_txt $(CFLAGS)
+distance_fast : distance_fast.c
+	$(CC) distance_fast.c -o distance_fast $(CFLAGS)
+kmeans_txt : kmeans_txt.c
+	$(CC) kmeans_txt.c -o kmeans_txt $(CFLAGS)
+word-analogy : word-analogy.c
+	$(CC) word-analogy.c -o word-analogy $(CFLAGS)
+compute-accuracy : compute-accuracy.c
+	$(CC) compute-accuracy.c -o compute-accuracy $(CFLAGS)
+clean:
+	rm -rf word2vec weightedWord2vec cngram2vec wiord2phrase distance word-analogy compute-accuracy distance_txt kmeans_txt
diff --git a/weightedWord2vec.c b/weightedWord2vec.c
new file mode 100644
index 0000000..f1d8f60
--- /dev/null
+++ b/weightedWord2vec.c
@@ -0,0 +1,1317 @@
+//  Copyright 2013 Google Inc. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <pthread.h>
+
+#define MAX_STRING 100
+#define EXP_TABLE_SIZE 1000
+#define MAX_EXP 6
+#define MAX_SENTENCE_LENGTH 1000
+#define MAX_CODE_LENGTH 40
+
+const int vocab_hash_size = 30000000;  // Maximum 30 * 0.7 = 21M words in the vocabulary
+
+typedef float real;                    // Precision of float numbers
+
+struct vocab_word {
+  long long cn;
+  int *point;
+  char *word, *code, codelen;
+};
+
+char train_file[MAX_STRING], output_file[MAX_STRING];
+char save_vocab_file[MAX_STRING], read_vocab_file[MAX_STRING];
+struct vocab_word *vocab;
+int binary = 0, type = 1, debug_mode = 2, window = 5, min_count = 5, num_threads = 12, min_reduce = 1;
+int *vocab_hash;
+long long vocab_max_size = 1000, vocab_size = 0, layer1_size = 100;
+long long train_words = 0, word_count_actual = 0, iter = 5, file_size = 0, classes = 0;
+real alpha = 0.025, starting_alpha, sample = 1e-3;
+real *syn0, *syn1, *syn1neg, *syn1nce, *expTable;
+clock_t start;
+
+real *syn1_window, *syn1neg_window, *syn1nce_window;
+int w_offset, window_layer_size;
+
+int window_hidden_size = 500; 
+real *syn_window_hidden, *syn_hidden_word, *syn_hidden_word_neg, *syn_hidden_word_nce; 
+
+int hs = 0, negative = 5;
+const int table_size = 1e8;
+int *table;
+
+//constrastive negative sampling
+char negative_classes_file[MAX_STRING];
+int *word_to_group;
+int *group_to_table; //group_size*table_size
+int class_number;
+
+//nce
+real* noise_distribution;
+int nce = 0;
+
+//param caps
+real CAP_VALUE = 50;
+int cap = 0;
+
+void capParam(real* array, int index){
+	if(array[index] > CAP_VALUE) 
+		array[index] = CAP_VALUE;
+	else if(array[index] < -CAP_VALUE)
+		array[index] = -CAP_VALUE; 
+}
+
+real hardTanh(real x){
+	if(x>=1){
+		return 1;
+	}
+	else if(x<=-1){
+		return -1;
+	}
+	else{
+		return x;
+	}
+}
+
+real dHardTanh(real x, real g){
+	if(x > 1 && g > 0){
+		return 0;
+	}
+	if(x < -1 && g < 0){
+		return 0;
+	}
+	return 1;
+}
+
+int isEndOfSentence(char* word){
+    return strcmp("</s>", word) == 0;
+}
+
+void InitUnigramTable() {
+  int a, i;
+  long long train_words_pow = 0;
+  real d1, power = 0.75;
+  table = (int *)malloc(table_size * sizeof(int));
+  for (a = 0; a < vocab_size; a++) train_words_pow += pow(vocab[a].cn, power);
+  i = 0;
+  d1 = pow(vocab[i].cn, power) / (real)train_words_pow;
+  for (a = 0; a < table_size; a++) {
+    table[a] = i;
+    if (a / (real)table_size > d1) {
+      i++;
+      d1 += pow(vocab[i].cn, power) / (real)train_words_pow;
+    }
+    if (i >= vocab_size) i = vocab_size - 1;
+  }
+  
+  noise_distribution = (real *)calloc(vocab_size, sizeof(real));
+  for (a = 0; a < vocab_size; a++) noise_distribution[a] = pow(vocab[a].cn, power)/(real)train_words_pow;
+}
+
+// Reads a single word from a file, assuming space + tab + EOL to be word boundaries
+void ReadWord(char *word, FILE *fin) {
+  int a = 0, ch;
+  while (!feof(fin)) {
+    ch = fgetc(fin);
+    if (ch == 13) continue;
+    if ((ch == ' ') || (ch == '\t') || (ch == '\n')) {
+      if (a > 0) {
+        if (ch == '\n') ungetc(ch, fin);
+        break;
+      }
+      if (ch == '\n') {
+        strcpy(word, (char *)"</s>");
+        return;
+      } else continue;
+    }
+    word[a] = ch;
+    a++;
+    if (a >= MAX_STRING - 1) a--;   // Truncate too long words
+  }
+  word[a] = 0;
+}
+
+// Returns hash value of a word
+int GetWordHash(char *word) {
+  unsigned long long a, hash = 0;
+  for (a = 0; a < strlen(word); a++) hash = hash * 257 + word[a];
+  hash = hash % vocab_hash_size;
+  return hash;
+}
+
+// Returns position of a word in the vocabulary; if the word is not found, returns -1
+int SearchVocab(char *word) {
+  unsigned int hash = GetWordHash(word);
+  while (1) {
+    if (vocab_hash[hash] == -1) return -1;
+    if (!strcmp(word, vocab[vocab_hash[hash]].word)) return vocab_hash[hash];
+    hash = (hash + 1) % vocab_hash_size;
+  }
+  return -1;
+}
+
+// Reads a word and returns its index in the vocabulary
+int ReadWordIndex(FILE *fin) {
+  char word[MAX_STRING];
+  ReadWord(word, fin);
+  if (feof(fin)) return -1;
+  return SearchVocab(word);
+}
+
+// Adds a word to the vocabulary
+int AddWordToVocab(char *word) {
+  unsigned int hash, length = strlen(word) + 1;
+  if (length > MAX_STRING) length = MAX_STRING;
+  vocab[vocab_size].word = (char *)calloc(length, sizeof(char));
+  strcpy(vocab[vocab_size].word, word);
+  vocab[vocab_size].cn = 0;
+  vocab_size++;
+  // Reallocate memory if needed
+  if (vocab_size + 2 >= vocab_max_size) {
+    vocab_max_size += 1000;
+    vocab = (struct vocab_word *)realloc(vocab, vocab_max_size * sizeof(struct vocab_word));
+  }
+  hash = GetWordHash(word);
+  while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+  vocab_hash[hash] = vocab_size - 1;
+  return vocab_size - 1;
+}
+
+// Used later for sorting by word counts
+int VocabCompare(const void *a, const void *b) {
+    return ((struct vocab_word *)b)->cn - ((struct vocab_word *)a)->cn;
+}
+
+// Sorts the vocabulary by frequency using word counts
+void SortVocab() {
+  int a, size;
+  unsigned int hash;
+  // Sort the vocabulary and keep </s> at the first position
+  qsort(&vocab[1], vocab_size - 1, sizeof(struct vocab_word), VocabCompare);
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  size = vocab_size;
+  train_words = 0;
+  for (a = 0; a < size; a++) {
+    // Words occuring less than min_count times will be discarded from the vocab
+    if ((vocab[a].cn < min_count) && (a != 0)) {
+      vocab_size--;
+      free(vocab[a].word);
+    } else {
+      // Hash will be re-computed, as after the sorting it is not actual
+      hash=GetWordHash(vocab[a].word);
+      while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+      vocab_hash[hash] = a;
+      train_words += vocab[a].cn;
+    }
+  }
+  vocab = (struct vocab_word *)realloc(vocab, (vocab_size + 1) * sizeof(struct vocab_word));
+  // Allocate memory for the binary tree construction
+  for (a = 0; a < vocab_size; a++) {
+    vocab[a].code = (char *)calloc(MAX_CODE_LENGTH, sizeof(char));
+    vocab[a].point = (int *)calloc(MAX_CODE_LENGTH, sizeof(int));
+  }
+}
+
+// Reduces the vocabulary by removing infrequent tokens
+void ReduceVocab() {
+  int a, b = 0;
+  unsigned int hash;
+  for (a = 0; a < vocab_size; a++) if (vocab[a].cn > min_reduce) {
+    vocab[b].cn = vocab[a].cn;
+    vocab[b].word = vocab[a].word;
+    b++;
+  } else free(vocab[a].word);
+  vocab_size = b;
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  for (a = 0; a < vocab_size; a++) {
+    // Hash will be re-computed, as it is not actual
+    hash = GetWordHash(vocab[a].word);
+    while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+    vocab_hash[hash] = a;
+  }
+  fflush(stdout);
+  min_reduce++;
+}
+
+// Create binary Huffman tree using the word counts
+// Frequent words will have short uniqe binary codes
+void CreateBinaryTree() {
+  long long a, b, i, min1i, min2i, pos1, pos2, point[MAX_CODE_LENGTH];
+  char code[MAX_CODE_LENGTH];
+  long long *count = (long long *)calloc(vocab_size * 2 + 1, sizeof(long long));
+  long long *binary = (long long *)calloc(vocab_size * 2 + 1, sizeof(long long));
+  long long *parent_node = (long long *)calloc(vocab_size * 2 + 1, sizeof(long long));
+  for (a = 0; a < vocab_size; a++) count[a] = vocab[a].cn;
+  for (a = vocab_size; a < vocab_size * 2; a++) count[a] = 1e15;
+  pos1 = vocab_size - 1;
+  pos2 = vocab_size;
+  // Following algorithm constructs the Huffman tree by adding one node at a time
+  for (a = 0; a < vocab_size - 1; a++) {
+    // First, find two smallest nodes 'min1, min2'
+    if (pos1 >= 0) {
+      if (count[pos1] < count[pos2]) {
+        min1i = pos1;
+        pos1--;
+      } else {
+        min1i = pos2;
+        pos2++;
+      }
+    } else {
+      min1i = pos2;
+      pos2++;
+    }
+    if (pos1 >= 0) {
+      if (count[pos1] < count[pos2]) {
+        min2i = pos1;
+        pos1--;
+      } else {
+        min2i = pos2;
+        pos2++;
+      }
+    } else {
+      min2i = pos2;
+      pos2++;
+    }
+    count[vocab_size + a] = count[min1i] + count[min2i];
+    parent_node[min1i] = vocab_size + a;
+    parent_node[min2i] = vocab_size + a;
+    binary[min2i] = 1;
+  }
+  // Now assign binary code to each vocabulary word
+  for (a = 0; a < vocab_size; a++) {
+    b = a;
+    i = 0;
+    while (1) {
+      code[i] = binary[b];
+      point[i] = b;
+      i++;
+      b = parent_node[b];
+      if (b == vocab_size * 2 - 2) break;
+    }
+    vocab[a].codelen = i;
+    vocab[a].point[0] = vocab_size - 2;
+    for (b = 0; b < i; b++) {
+      vocab[a].code[i - b - 1] = code[b];
+      vocab[a].point[i - b] = point[b] - vocab_size;
+    }
+  }
+  free(count);
+  free(binary);
+  free(parent_node);
+}
+
+void LearnVocabFromTrainFile() {
+  char word[MAX_STRING];
+  FILE *fin;
+  long long a, i;
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  fin = fopen(train_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: training data file not found!\n");
+    exit(1);
+  }
+  vocab_size = 0;
+  AddWordToVocab((char *)"</s>");
+  int startOfLine = 1;
+  while (1) {
+    ReadWord(word, fin);
+    if (feof(fin)) break;
+    if (startOfLine) {
+      ReadWord(word, fin);
+      startOfLine = 0;
+    }
+    if(isEndOfSentence(word)){
+      startOfLine = 1;
+    }
+    train_words++;
+    if ((debug_mode > 1) && (train_words % 100000 == 0)) {
+      printf("%lldK%c", train_words / 1000, 13);
+      fflush(stdout);
+    }
+    i = SearchVocab(word);
+    if (i == -1) {
+      a = AddWordToVocab(word);
+      vocab[a].cn = 1;
+    } else vocab[i].cn++;
+    if (vocab_size > vocab_hash_size * 0.7) ReduceVocab();
+  }
+  SortVocab();
+  if (debug_mode > 0) {
+    printf("Vocab size: %lld\n", vocab_size);
+    printf("Words in train file: %lld\n", train_words);
+  }
+  file_size = ftell(fin);
+  fclose(fin);
+}
+
+void SaveVocab() {
+  long long i;
+  FILE *fo = fopen(save_vocab_file, "wb");
+  for (i = 0; i < vocab_size; i++) fprintf(fo, "%s %lld\n", vocab[i].word, vocab[i].cn);
+  fclose(fo);
+}
+
+void ReadVocab() {
+  long long a, i = 0;
+  char c;
+  char word[MAX_STRING];
+  FILE *fin = fopen(read_vocab_file, "rb");
+  if (fin == NULL) {
+    printf("Vocabulary file not found\n");
+    exit(1);
+  }
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  vocab_size = 0;
+  while (1) {
+    ReadWord(word, fin);
+    if (feof(fin)) break;
+    a = AddWordToVocab(word);
+    fscanf(fin, "%lld%c", &vocab[a].cn, &c);
+    i++;
+  }
+  SortVocab();
+  if (debug_mode > 0) {
+    printf("Vocab size: %lld\n", vocab_size);
+    printf("Words in train file: %lld\n", train_words);
+  }
+  fin = fopen(train_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: training data file not found!\n");
+    exit(1);
+  }
+  fseek(fin, 0, SEEK_END);
+  file_size = ftell(fin);
+  fclose(fin);
+}
+
+void InitClassUnigramTable() {
+  long long a,c;
+  printf("loading class unigrams \n");
+  FILE *fin = fopen(negative_classes_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: class file not found!\n");
+    exit(1);
+  }
+  word_to_group = (int *)malloc(vocab_size * sizeof(int));
+  for(a = 0; a < vocab_size; a++) word_to_group[a] = -1;
+  char class[MAX_STRING];
+  char prev_class[MAX_STRING];
+  prev_class[0] = 0;
+  char word[MAX_STRING];
+  class_number = -1;
+  while (1) {
+    if (feof(fin)) break;
+    ReadWord(class, fin);
+    ReadWord(word, fin);
+    int word_index = SearchVocab(word);
+    if (word_index != -1){
+       if(strcmp(class, prev_class) != 0){
+	    class_number++;
+	    strcpy(prev_class, class);
+       }
+       word_to_group[word_index] = class_number;
+    }
+    ReadWord(word, fin);
+  }
+  class_number++;
+  fclose(fin);
+  
+  group_to_table = (int *)malloc(table_size * class_number * sizeof(int)); 
+  long long train_words_pow = 0;
+  real d1, power = 0.75;
+  
+  for(c = 0; c < class_number; c++){
+     long long offset = c * table_size;
+     train_words_pow = 0;
+     for (a = 0; a < vocab_size; a++) if(word_to_group[a] == c) train_words_pow += pow(vocab[a].cn, power);
+     int i = 0;
+     while(word_to_group[i]!=c && i < vocab_size) i++;
+     d1 = pow(vocab[i].cn, power) / (real)train_words_pow;
+     for (a = 0; a < table_size; a++) {
+	//printf("index %lld , word %d\n", a, i);
+	group_to_table[offset + a] = i;
+        if (a / (real)table_size > d1) {
+	   i++;
+           while(word_to_group[i]!=c && i < vocab_size) i++;
+	   d1 += pow(vocab[i].cn, power) / (real)train_words_pow;
+        }
+        if (i >= vocab_size) while(word_to_group[i]!=c && i >= 0) i--;
+     }
+  }
+}
+
+void InitNet() {
+  long long a, b;
+  unsigned long long next_random = 1;
+  window_layer_size = layer1_size*window*2;
+  a = posix_memalign((void **)&syn0, 128, (long long)vocab_size * layer1_size * sizeof(real));
+  if (syn0 == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  
+  if (hs) {
+    a = posix_memalign((void **)&syn1, 128, (long long)vocab_size * layer1_size * sizeof(real));
+    if (syn1 == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn1_window, 128, (long long)vocab_size * window_layer_size * sizeof(real));
+    if (syn1_window == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn_hidden_word, 128, (long long)vocab_size * window_hidden_size * sizeof(real));
+    if (syn_hidden_word == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++)
+     syn1[a * layer1_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_layer_size; b++)
+     syn1_window[a * window_layer_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_hidden_size; b++)
+     syn_hidden_word[a * window_hidden_size + b] = 0;
+  }
+  if (negative>0) {
+    a = posix_memalign((void **)&syn1neg, 128, (long long)vocab_size * layer1_size * sizeof(real));
+    if (syn1neg == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn1neg_window, 128, (long long)vocab_size * window_layer_size * sizeof(real));
+    if (syn1neg_window == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn_hidden_word_neg, 128, (long long)vocab_size * window_hidden_size * sizeof(real));
+    if (syn_hidden_word_neg == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++)
+     syn1neg[a * layer1_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_layer_size; b++)
+     syn1neg_window[a * window_layer_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_hidden_size; b++)
+     syn_hidden_word_neg[a * window_hidden_size + b] = 0;
+  }
+  if (nce>0) {
+    a = posix_memalign((void **)&syn1nce, 128, (long long)vocab_size * layer1_size * sizeof(real));
+    if (syn1nce == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn1nce_window, 128, (long long)vocab_size * window_layer_size * sizeof(real));
+    if (syn1nce_window == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn_hidden_word_nce, 128, (long long)vocab_size * window_hidden_size * sizeof(real));
+    if (syn_hidden_word_nce == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++)
+     syn1nce[a * layer1_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_layer_size; b++)
+     syn1nce_window[a * window_layer_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_hidden_size; b++)
+     syn_hidden_word_nce[a * window_hidden_size + b] = 0;
+  }
+  for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++) {
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    syn0[a * layer1_size + b] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / layer1_size;
+  }
+
+  a = posix_memalign((void **)&syn_window_hidden, 128, window_hidden_size * window_layer_size * sizeof(real));
+  if (syn_window_hidden == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  for (a = 0; a < window_hidden_size * window_layer_size; a++){
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    syn_window_hidden[a] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / (window_hidden_size*window_layer_size);
+  }
+
+  CreateBinaryTree();
+}
+
+long long findStartOfLine(char* file, long long start){
+  char word[MAX_STRING];
+  if(start == 0) return 0;
+  while(start != 0){
+    FILE*fi = fopen(file, "rb");
+    fseek(fi, start, SEEK_SET);
+    ReadWord(word, fi);
+    if(isEndOfSentence(word)){
+      fclose(fi);
+      return start+1;
+    }
+    fclose(fi);
+    start--;
+  }
+  return 0;
+}
+
+void *TrainModelThread(void *id) {
+  char word_str[MAX_STRING];
+  long long a, b, d, cw, word, last_word, sentence_length = 0, sentence_position = 0;
+  long long word_count = 0, last_word_count = 0, sen[MAX_SENTENCE_LENGTH + 1];
+  long long l1, l2, c, target, label, local_iter = iter;
+  unsigned long long next_random = (long long)id;
+  real f, g;
+  clock_t now;
+  int input_len_1 = layer1_size;
+  int window_offset = -1;
+  float currentWeight = 0;
+  if(type == 2 || type == 4){
+     input_len_1=window_layer_size;
+  }
+  real *neu1 = (real *)calloc(input_len_1, sizeof(real));
+  real *neu1e = (real *)calloc(input_len_1, sizeof(real));
+
+  int input_len_2 = 0;
+  if(type == 4){
+     input_len_2 = window_hidden_size;
+  }
+  real *neu2 = (real *)calloc(input_len_2, sizeof(real));
+  real *neu2e = (real *)calloc(input_len_2, sizeof(real));
+
+  long long start_pos = findStartOfLine(train_file, file_size / (long long)num_threads * (long long)id);
+  FILE *fi = fopen(train_file, "rb");
+  fseek(fi, start_pos, SEEK_SET);
+  int startOfSentence = 1;
+  int startEndOfLineIndex = SearchVocab("</s>");
+  while (1) {
+    if (word_count - last_word_count > 10000) {
+      word_count_actual += word_count - last_word_count;
+      last_word_count = word_count;
+      if ((debug_mode > 1)) {
+        now=clock();
+        printf("%cAlpha: %f Weight: %f  Progress: %.2f%%  Words/thread/sec: %.2fk  ", 13, alpha, currentWeight,
+         word_count_actual / (real)(iter * train_words + 1) * 100,
+         word_count_actual / ((real)(now - start + 1) / (real)CLOCKS_PER_SEC * 1000));
+        fflush(stdout);
+      }
+      alpha = starting_alpha * (1 - word_count_actual / (real)(iter * train_words + 1));
+      if (alpha < starting_alpha * 0.0001) alpha = starting_alpha * 0.0001;
+    }
+    if (sentence_length == 0) {
+      while (1) {
+        if(startOfSentence){
+          ReadWord(word_str, fi);
+          currentWeight = atof(word_str);
+          startOfSentence = 0;
+          continue;
+        }
+        word = ReadWordIndex(fi);
+        if (word == startEndOfLineIndex){
+          startOfSentence = 1;
+        }
+        if (feof(fi)) break;
+        if (word == -1) continue;
+        word_count++;
+        if (word == 0) break;
+        // The subsampling randomly discards frequent words while keeping the ranking same
+        if (sample > 0) {
+          real ran = (sqrt(vocab[word].cn / (sample * train_words)) + 1) * (sample * train_words) / vocab[word].cn;
+          next_random = next_random * (unsigned long long)25214903917 + 11;
+          if (ran < (next_random & 0xFFFF) / (real)65536) continue;
+        }
+        sen[sentence_length] = word;
+        sentence_length++;
+        if (sentence_length >= MAX_SENTENCE_LENGTH) break;
+      }
+      sentence_position = 0;
+    }
+    if (feof(fi) || (word_count > train_words / num_threads)) {
+      word_count_actual += word_count - last_word_count;
+      local_iter--;
+      if (local_iter == 0) break;
+      word_count = 0;
+      last_word_count = 0;
+      sentence_length = 0;
+      fseek(fi, start_pos, SEEK_SET);
+      continue;
+    }
+    word = sen[sentence_position];
+    if (word == -1) continue;
+    for (c = 0; c < input_len_1; c++) neu1[c] = 0;
+    for (c = 0; c < input_len_1; c++) neu1e[c] = 0;
+    for (c = 0; c < input_len_2; c++) neu2[c] = 0;
+    for (c = 0; c < input_len_2; c++) neu2e[c] = 0;
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    b = next_random % window;
+    if (type == 0) {  //train the cbow architecture
+      // in -> hidden
+      cw = 0;
+      for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        for (c = 0; c < layer1_size; c++) neu1[c] += syn0[c + last_word * layer1_size];
+        cw++;
+      }
+      if (cw) {
+        for (c = 0; c < layer1_size; c++) neu1[c] /= cw;
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * layer1_size;
+          // Propagate hidden -> output
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha * currentWeight;
+          // Propagate errors output -> hidden
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < layer1_size; c++) syn1[c + l2] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1, c + l2);
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+	    if(word_to_group != NULL && word_to_group[word] != -1){
+		target = word;
+		while(target == word) {
+			target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+            		next_random = next_random * (unsigned long long)25214903917 + 11;
+		}
+		//printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+	    }
+	    else{
+            	target = table[(next_random >> 16) % table_size];
+	    }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1neg[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha * currentWeight;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha * currentWeight;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha * currentWeight;
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1neg[c + l2] += g * neu1[c];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1neg, c + l2);
+        }
+        // Noise Contrastive Estimation
+        if (nce > 0) for (d = 0; d < nce + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+	    if(word_to_group != NULL && word_to_group[word] != -1){
+		target = word;
+		while(target == word) {
+			target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+            		next_random = next_random * (unsigned long long)25214903917 + 11;
+		}
+	    }
+	    else{
+            	target = table[(next_random >> 16) % table_size];
+	    }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+	  
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1nce[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha * currentWeight;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha * currentWeight;
+          else {
+                f = exp(f);
+                g = (label - f/(noise_distribution[target]*nce + f)) * alpha * currentWeight;
+          }
+	  for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1nce[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1nce[c + l2] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1nce,c + l2);
+        }
+        // hidden -> in
+        for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {
+          c = sentence_position - window + a;
+          if (c < 0) continue;
+          if (c >= sentence_length) continue;
+          last_word = sen[c];
+          if (last_word == -1) continue;
+          for (c = 0; c < layer1_size; c++) syn0[c + last_word * layer1_size] += neu1e[c];
+        }
+      }
+    } else if(type==1) {  //train skip-gram
+      for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        l1 = last_word * layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1e[c] = 0;
+        // HIERARCHICAL SOFTMAX
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * layer1_size;
+          // Propagate hidden -> output
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha * currentWeight;
+          // Propagate errors output -> hidden
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < layer1_size; c++) syn1[c + l2] += g * syn0[c + l1];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1, c + l2);
+        }
+        // NEGATIVE SAMPLING
+	if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	    next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1neg[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha * currentWeight;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha * currentWeight;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha * currentWeight;
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1neg[c + l2] += g * syn0[c + l1];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1neg, c + l2);
+        }
+	//Noise Contrastive Estimation
+        if (nce > 0) for (d = 0; d < nce + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	    next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1nce[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha * currentWeight;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha * currentWeight;
+          else {
+                f = exp(f);
+                g = (label - f/(noise_distribution[target]*nce + f)) * alpha * currentWeight;
+          }
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1nce[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1nce[c + l2] += g * syn0[c + l1];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1nce, c + l2);
+        }
+        // Learn weights input -> hidden
+        for (c = 0; c < layer1_size; c++) syn0[c + l1] += neu1e[c];
+      }
+    }
+    else if(type == 2){ //train the cwindow architecture
+      // in -> hidden
+      cw = 0;
+      for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        window_offset = a*layer1_size;
+        if (a > window) window_offset-=layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1[c+window_offset] += syn0[c + last_word * layer1_size];
+        cw++;
+      }
+      if (cw) {
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * window_layer_size;
+          // Propagate hidden -> output
+          for (c = 0; c < window_layer_size; c++) f += neu1[c] * syn1_window[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha * currentWeight;
+          // Propagate errors output -> hidden
+          for (c = 0; c < window_layer_size; c++) neu1e[c] += g * syn1_window[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < window_layer_size; c++) syn1_window[c + l2] += g * neu1[c];
+	  if (cap == 1) for (c = 0; c < window_layer_size; c++) capParam(syn1_window, c + l2);
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < window_layer_size; c++) f += neu1[c] * syn1neg_window[c + l2];
+	  if (f > MAX_EXP) g = (label - 1) * alpha * currentWeight;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha * currentWeight;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha * currentWeight;
+          for (c = 0; c < window_layer_size; c++) neu1e[c] += g * syn1neg_window[c + l2];
+          for (c = 0; c < window_layer_size; c++) syn1neg_window[c + l2] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < window_layer_size; c++) capParam(syn1neg_window, c + l2);
+        }
+	// Noise Contrastive Estimation
+        if (nce > 0) for (d = 0; d < nce + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < window_layer_size; c++) f += neu1[c] * syn1nce_window[c + l2];
+	  if (f > MAX_EXP) g = (label - 1) * alpha * currentWeight;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha * currentWeight;
+          else {
+                f = exp(f);
+                g = (label - f/(noise_distribution[target]*nce + f)) * alpha * currentWeight;
+          }
+          for (c = 0; c < window_layer_size; c++) neu1e[c] += g * syn1nce_window[c + l2];
+          for (c = 0; c < window_layer_size; c++) syn1nce_window[c + l2] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < window_layer_size; c++) capParam(syn1nce_window, c + l2);
+        }
+        // hidden -> in
+        for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+          c = sentence_position - window + a;
+          if (c < 0) continue;
+          if (c >= sentence_length) continue;
+          last_word = sen[c];
+          if (last_word == -1) continue;
+	  window_offset = a * layer1_size;
+	  if(a > window) window_offset -= layer1_size;
+          for (c = 0; c < layer1_size; c++) syn0[c + last_word * layer1_size] += neu1e[c + window_offset];
+        }
+      }
+    }
+    else if (type == 3){  //train structured skip-gram
+      for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        l1 = last_word * layer1_size;
+	window_offset = a * layer1_size;
+	if(a > window) window_offset -= layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1e[c] = 0;
+        // HIERARCHICAL SOFTMAX
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * window_layer_size;
+          // Propagate hidden -> output
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1_window[c + l2 + window_offset];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha * currentWeight;
+          // Propagate errors output -> hidden
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1_window[c + l2 + window_offset];
+          // Learn weights hidden -> output
+          for (c = 0; c < layer1_size; c++) syn1[c + l2 + window_offset] += g * syn0[c + l1];
+	  if(cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1, c + l2 + window_offset);
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	     next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1neg_window[c + l2 + window_offset];
+	  if (f > MAX_EXP) g = (label - 1) * alpha * currentWeight;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha * currentWeight;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha * currentWeight;
+	  for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg_window[c + l2 + window_offset];
+          for (c = 0; c < layer1_size; c++) syn1neg_window[c + l2 + window_offset] += g * syn0[c + l1]; 
+	  if(cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1neg_window, c + l2 + window_offset);
+        }
+	// Noise Constrastive Estimation
+        if (nce > 0) for (d = 0; d < nce + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+             next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1nce_window[c + l2 + window_offset];
+	  if (f > MAX_EXP) g = (label - 1) * alpha * currentWeight;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha * currentWeight;
+          else {
+                f = exp(f);
+                g = (label - f/(noise_distribution[target]*nce + f)) * alpha * currentWeight;
+          }
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1nce_window[c + l2 + window_offset];
+          for (c = 0; c < layer1_size; c++) syn1nce_window[c + l2 + window_offset] += g * syn0[c + l1];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1nce_window, c + l2 + window_offset);
+        }
+        // Learn weights input -> hidden
+        for (c = 0; c < layer1_size; c++) {syn0[c + l1] += neu1e[c]; if(syn0[c + l1] > 50) syn0[c + l1] = 50; if(syn0[c + l1] < -50) syn0[c + l1] = -50;}
+      }
+    }
+    else if(type == 4){ //training senna
+	// in -> hidden
+      cw = 0;
+      for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        window_offset = a*layer1_size;
+        if (a > window) window_offset-=layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1[c+window_offset] += syn0[c + last_word * layer1_size];
+        cw++;
+      }
+      if (cw) {
+		for (a = 0; a < window_hidden_size; a++){
+          c = a*window_layer_size;
+          for(b = 0; b < window_layer_size; b++){
+             neu2[a] += syn_window_hidden[c + b] * neu1[b];
+          }
+        }
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * window_hidden_size;
+          // Propagate hidden -> output
+          for (c = 0; c < window_hidden_size; c++) f += hardTanh(neu2[c]) * syn_hidden_word[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha * currentWeight;
+          // Propagate errors output -> hidden
+          for (c = 0; c < window_hidden_size; c++) neu2e[c] += dHardTanh(neu2[c],g) * g * syn_hidden_word[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < window_hidden_size; c++) syn_hidden_word[c + l2] += dHardTanh(neu2[c],g) * g * neu2[c];
+        }
+      // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	    next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_hidden_size;
+          f = 0;
+          for (c = 0; c < window_hidden_size; c++) f += hardTanh(neu2[c]) * syn_hidden_word_neg[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha  * currentWeight / negative;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha * currentWeight / negative;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha * currentWeight / negative;
+          for (c = 0; c < window_hidden_size; c++) neu2e[c] += dHardTanh(neu2[c],g) * g * syn_hidden_word_neg[c + l2];
+          for (c = 0; c < window_hidden_size; c++) syn_hidden_word_neg[c + l2] += dHardTanh(neu2[c],g) * g * neu2[c];
+        }
+        for (a = 0; a < window_hidden_size; a++)
+          for(b = 0; b < window_layer_size; b++)
+	     neu1e[b] += neu2e[a] * syn_window_hidden[a*window_layer_size + b];
+	for (a = 0; a < window_hidden_size; a++)
+          for(b = 0; b < window_layer_size; b++)
+	     syn_window_hidden[a*window_layer_size + b] += neu2e[a] * neu1[b];
+        // hidden -> in
+        for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+          c = sentence_position - window + a;
+          if (c < 0) continue;
+          if (c >= sentence_length) continue;
+          last_word = sen[c];
+          if (last_word == -1) continue;
+          window_offset = a * layer1_size;
+          if(a > window) window_offset -= layer1_size;
+          for (c = 0; c < layer1_size; c++) syn0[c + last_word * layer1_size] += neu1e[c + window_offset];
+        }
+      }
+    }
+    else{
+	printf("unknown type %i", type);
+	exit(0);
+    }
+    sentence_position++;
+    if (sentence_position >= sentence_length) {
+      sentence_length = 0;
+      continue;
+    }
+  }
+  fclose(fi);
+  free(neu1);
+  free(neu1e);
+  pthread_exit(NULL);
+}
+
+void TrainModel() {
+  long a, b, c, d;
+  FILE *fo;
+  pthread_t *pt = (pthread_t *)malloc(num_threads * sizeof(pthread_t));
+  printf("Starting training using file %s\n", train_file);
+  starting_alpha = alpha;
+  if (read_vocab_file[0] != 0) ReadVocab(); else LearnVocabFromTrainFile();
+  if (save_vocab_file[0] != 0) SaveVocab();
+  if (output_file[0] == 0) return;
+  InitNet();
+  if (negative > 0 || nce > 0) InitUnigramTable();
+  if (negative_classes_file[0] != 0) InitClassUnigramTable();
+  start = clock();
+  for (a = 0; a < num_threads; a++) pthread_create(&pt[a], NULL, TrainModelThread, (void *)a);
+  for (a = 0; a < num_threads; a++) pthread_join(pt[a], NULL);
+  fo = fopen(output_file, "wb");
+  if (classes == 0) {
+    // Save the word vectors
+    fprintf(fo, "%lld %lld\n", vocab_size, layer1_size);
+    for (a = 0; a < vocab_size; a++) {
+      fprintf(fo, "%s ", vocab[a].word);
+      if (binary) for (b = 0; b < layer1_size; b++) fwrite(&syn0[a * layer1_size + b], sizeof(real), 1, fo);
+      else for (b = 0; b < layer1_size; b++) fprintf(fo, "%lf ", syn0[a * layer1_size + b]);
+      fprintf(fo, "\n");
+    }
+  } else {
+    // Run K-means on the word vectors
+    int clcn = classes, iter = 10, closeid;
+    int *centcn = (int *)malloc(classes * sizeof(int));
+    int *cl = (int *)calloc(vocab_size, sizeof(int));
+    real closev, x;
+    real *cent = (real *)calloc(classes * layer1_size, sizeof(real));
+    for (a = 0; a < vocab_size; a++) cl[a] = a % clcn;
+    for (a = 0; a < iter; a++) {
+      for (b = 0; b < clcn * layer1_size; b++) cent[b] = 0;
+      for (b = 0; b < clcn; b++) centcn[b] = 1;
+      for (c = 0; c < vocab_size; c++) {
+        for (d = 0; d < layer1_size; d++) cent[layer1_size * cl[c] + d] += syn0[c * layer1_size + d];
+        centcn[cl[c]]++;
+      }
+      for (b = 0; b < clcn; b++) {
+        closev = 0;
+        for (c = 0; c < layer1_size; c++) {
+          cent[layer1_size * b + c] /= centcn[b];
+          closev += cent[layer1_size * b + c] * cent[layer1_size * b + c];
+        }
+        closev = sqrt(closev);
+        for (c = 0; c < layer1_size; c++) cent[layer1_size * b + c] /= closev;
+      }
+      for (c = 0; c < vocab_size; c++) {
+        closev = -10;
+        closeid = 0;
+        for (d = 0; d < clcn; d++) {
+          x = 0;
+          for (b = 0; b < layer1_size; b++) x += cent[layer1_size * d + b] * syn0[c * layer1_size + b];
+          if (x > closev) {
+            closev = x;
+            closeid = d;
+          }
+        }
+        cl[c] = closeid;
+      }
+    }
+    // Save the K-means classes
+    for (a = 0; a < vocab_size; a++) fprintf(fo, "%s %d\n", vocab[a].word, cl[a]);
+    free(centcn);
+    free(cent);
+    free(cl);
+  }
+  fclose(fo);
+}
+
+int ArgPos(char *str, int argc, char **argv) {
+  int a;
+  for (a = 1; a < argc; a++) if (!strcmp(str, argv[a])) {
+    if (a == argc - 1) {
+      printf("Argument missing for %s\n", str);
+      exit(1);
+    }
+    return a;
+  }
+  return -1;
+}
+
+int main(int argc, char **argv) {
+  int i;
+  if (argc == 1) {
+    printf("WORD VECTOR estimation toolkit v 0.1c\n\n");
+    printf("Options:\n");
+    printf("Parameters for training:\n");
+    printf("\t-train <file>\n");
+    printf("\t\tUse text data from <file> to train the model\n");
+    printf("\t-output <file>\n");
+    printf("\t\tUse <file> to save the resulting word vectors / word clusters\n");
+    printf("\t-size <int>\n");
+    printf("\t\tSet size of word vectors; default is 100\n");
+    printf("\t-window <int>\n");
+    printf("\t\tSet max skip length between words; default is 5\n");
+    printf("\t-sample <float>\n");
+    printf("\t\tSet threshold for occurrence of words. Those that appear with higher frequency in the training data\n");
+    printf("\t\twill be randomly down-sampled; default is 1e-3, useful range is (0, 1e-5)\n");
+    printf("\t-hs <int>\n");
+    printf("\t\tUse Hierarchical Softmax; default is 0 (not used)\n");
+    printf("\t-negative <int>\n");
+    printf("\t\tNumber of negative examples; default is 5, common values are 3 - 10 (0 = not used)\n");
+    printf("\t-negative-classes <file>\n");
+    printf("\t\tNegative classes to sample from\n");
+    printf("\t-nce <int>\n");
+    printf("\t\tNumber of negative examples for nce; default is 0, common values are 3 - 10 (0 = not used)\n");
+    printf("\t-threads <int>\n");
+    printf("\t\tUse <int> threads (default 12)\n");
+    printf("\t-iter <int>\n");
+    printf("\t\tRun more training iterations (default 5)\n");
+    printf("\t-min-count <int>\n");
+    printf("\t\tThis will discard words that appear less than <int> times; default is 5\n");
+    printf("\t-alpha <float>\n");
+    printf("\t\tSet the starting learning rate; default is 0.025 for skip-gram and 0.05 for CBOW\n");
+    printf("\t-classes <int>\n");
+    printf("\t\tOutput word classes rather than word vectors; default number of classes is 0 (vectors are written)\n");
+    printf("\t-debug <int>\n");
+    printf("\t\tSet the debug mode (default = 2 = more info during training)\n");
+    printf("\t-binary <int>\n");
+    printf("\t\tSave the resulting vectors in binary moded; default is 0 (off)\n");
+    printf("\t-save-vocab <file>\n");
+    printf("\t\tThe vocabulary will be saved to <file>\n");
+    printf("\t-read-vocab <file>\n");
+    printf("\t\tThe vocabulary will be read from <file>, not constructed from the training data\n");
+    printf("\t-type <int>\n");
+    printf("\t\tType of embeddings (0 for cbow, 1 for skipngram, 2 for cwindow, 3 for structured skipngram, 4 for senna type)\n");
+    printf("\t-cap <int>\n");
+    printf("\t\tlimit the parameter values to the range [-50, 50]; default is 0 (off)\n");
+    printf("\nExamples:\n");
+    printf("./word2vec -train data.txt -output vec.txt -size 200 -window 5 -sample 1e-4 -negative 5 -hs 0 -binary 0 -type 1 -iter 3\n\n");
+    return 0;
+  }
+  output_file[0] = 0;
+  save_vocab_file[0] = 0;
+  read_vocab_file[0] = 0;
+  negative_classes_file[0] = 0;
+  if ((i = ArgPos((char *)"-size", argc, argv)) > 0) layer1_size = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-train", argc, argv)) > 0) strcpy(train_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-save-vocab", argc, argv)) > 0) strcpy(save_vocab_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-read-vocab", argc, argv)) > 0) strcpy(read_vocab_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-debug", argc, argv)) > 0) debug_mode = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-binary", argc, argv)) > 0) binary = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-type", argc, argv)) > 0) type = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-output", argc, argv)) > 0) strcpy(output_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-window", argc, argv)) > 0) window = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-sample", argc, argv)) > 0) sample = atof(argv[i + 1]);
+  if ((i = ArgPos((char *)"-hs", argc, argv)) > 0) hs = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-negative", argc, argv)) > 0) negative = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-negative-classes", argc, argv)) > 0) strcpy(negative_classes_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-nce", argc, argv)) > 0) nce = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-threads", argc, argv)) > 0) num_threads = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-iter", argc, argv)) > 0) iter = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-min-count", argc, argv)) > 0) min_count = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-classes", argc, argv)) > 0) classes = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-cap", argc, argv)) > 0) cap = atoi(argv[i + 1]);
+  if (type==0 || type==2 || type==4) alpha = 0.05;
+  if ((i = ArgPos((char *)"-alpha", argc, argv)) > 0) alpha = atof(argv[i + 1]);
+  vocab = (struct vocab_word *)calloc(vocab_max_size, sizeof(struct vocab_word));
+  vocab_hash = (int *)calloc(vocab_hash_size, sizeof(int));
+  expTable = (real *)malloc((EXP_TABLE_SIZE + 1) * sizeof(real));
+  for (i = 0; i < EXP_TABLE_SIZE; i++) {
+    expTable[i] = exp((i / (real)EXP_TABLE_SIZE * 2 - 1) * MAX_EXP); // Precompute the exp() table
+    expTable[i] = expTable[i] / (expTable[i] + 1);                   // Precompute f(x) = x / (x + 1)
+  }
+  TrainModel();
+  return 0;
+}
+
diff --git a/word-analogy.c b/word-analogy.c
new file mode 100644
index 0000000..bc78ba7
--- /dev/null
+++ b/word-analogy.c
@@ -0,0 +1,143 @@
+//  Copyright 2013 Google Inc. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+
+#include <stdio.h>
+#include <string.h>
+#include <math.h>
+#include <stdlib.h>
+
+const long long max_size = 2000;         // max length of strings
+const long long N = 40;                  // number of closest words that will be shown
+const long long max_w = 50;              // max length of vocabulary entries
+
+int main(int argc, char **argv) {
+  FILE *f;
+  char st1[max_size];
+  char bestw[N][max_size];
+  char file_name[max_size], st[100][max_size];
+  float dist, len, bestd[N], vec[max_size];
+  long long words, size, a, b, c, d, cn, bi[100];
+  float *M;
+  char *vocab;
+  if (argc < 2) {
+    printf("Usage: ./word-analogy <FILE>\nwhere FILE contains word projections in the BINARY FORMAT\n");
+    return 0;
+  }
+  strcpy(file_name, argv[1]);
+  f = fopen(file_name, "rb");
+  if (f == NULL) {
+    printf("Input file not found\n");
+    return -1;
+  }
+  fscanf(f, "%lld", &words);
+  fscanf(f, "%lld", &size);
+  vocab = (char *)malloc((long long)words * max_w * sizeof(char));
+  M = (float *)malloc((long long)words * (long long)size * sizeof(float));
+  if (M == NULL) {
+    printf("Cannot allocate memory: %lld MB    %lld  %lld\n", (long long)words * size * sizeof(float) / 1048576, words, size);
+    return -1;
+  }
+  for (b = 0; b < words; b++) {
+    a = 0;
+    while (1) {
+      vocab[b * max_w + a] = fgetc(f);
+      if (feof(f) || (vocab[b * max_w + a] == ' ')) break;
+      if ((a < max_w) && (vocab[b * max_w + a] != '\n')) a++;
+    }
+    vocab[b * max_w + a] = 0;
+    for (a = 0; a < size; a++) fread(&M[a + b * size], sizeof(float), 1, f);
+    len = 0;
+    for (a = 0; a < size; a++) len += M[a + b * size] * M[a + b * size];
+    len = sqrt(len);
+    for (a = 0; a < size; a++) M[a + b * size] /= len;
+  }
+  fclose(f);
+  while (1) {
+    for (a = 0; a < N; a++) bestd[a] = 0;
+    for (a = 0; a < N; a++) bestw[a][0] = 0;
+    printf("Enter three words (EXIT to break): ");
+    a = 0;
+    while (1) {
+      st1[a] = fgetc(stdin);
+      if ((st1[a] == '\n') || (a >= max_size - 1)) {
+        st1[a] = 0;
+        break;
+      }
+      a++;
+    }
+    if (!strcmp(st1, "EXIT")) break;
+    cn = 0;
+    b = 0;
+    c = 0;
+    while (1) {
+      st[cn][b] = st1[c];
+      b++;
+      c++;
+      st[cn][b] = 0;
+      if (st1[c] == 0) break;
+      if (st1[c] == ' ') {
+        cn++;
+        b = 0;
+        c++;
+      }
+    }
+    cn++;
+    if (cn < 3) {
+      printf("Only %lld words were entered.. three words are needed at the input to perform the calculation\n", cn);
+      continue;
+    }
+    for (a = 0; a < cn; a++) {
+      for (b = 0; b < words; b++) if (!strcmp(&vocab[b * max_w], st[a])) break;
+      if (b == words) b = 0;
+      bi[a] = b;
+      printf("\nWord: %s  Position in vocabulary: %lld\n", st[a], bi[a]);
+      if (b == 0) {
+        printf("Out of dictionary word!\n");
+        break;
+      }
+    }
+    if (b == 0) continue;
+    printf("\n                                              Word              Distance\n------------------------------------------------------------------------\n");
+    for (a = 0; a < size; a++) vec[a] = M[a + bi[1] * size] - M[a + bi[0] * size] + M[a + bi[2] * size];
+    len = 0;
+    for (a = 0; a < size; a++) len += vec[a] * vec[a];
+    len = sqrt(len);
+    for (a = 0; a < size; a++) vec[a] /= len;
+    for (a = 0; a < N; a++) bestd[a] = 0;
+    for (a = 0; a < N; a++) bestw[a][0] = 0;
+    for (c = 0; c < words; c++) {
+      if (c == bi[0]) continue;
+      if (c == bi[1]) continue;
+      if (c == bi[2]) continue;
+      a = 0;
+      for (b = 0; b < cn; b++) if (bi[b] == c) a = 1;
+      if (a == 1) continue;
+      dist = 0;
+      for (a = 0; a < size; a++) dist += vec[a] * M[a + c * size];
+      for (a = 0; a < N; a++) {
+        if (dist > bestd[a]) {
+          for (d = N - 1; d > a; d--) {
+            bestd[d] = bestd[d - 1];
+            strcpy(bestw[d], bestw[d - 1]);
+          }
+          bestd[a] = dist;
+          strcpy(bestw[a], &vocab[c * max_w]);
+          break;
+        }
+      }
+    }
+    for (a = 0; a < N; a++) printf("%50s\t\t%f\n", bestw[a], bestd[a]);
+  }
+  return 0;
+}
diff --git a/word2phrase.c b/word2phrase.c
new file mode 100644
index 0000000..24238bc
--- /dev/null
+++ b/word2phrase.c
@@ -0,0 +1,292 @@
+//  Copyright 2013 Google Inc. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <pthread.h>
+
+#define MAX_STRING 60
+
+const int vocab_hash_size = 500000000; // Maximum 500M entries in the vocabulary
+
+typedef float real;                    // Precision of float numbers
+
+struct vocab_word {
+  long long cn;
+  char *word;
+};
+
+char train_file[MAX_STRING], output_file[MAX_STRING];
+struct vocab_word *vocab;
+int debug_mode = 2, min_count = 5, *vocab_hash, min_reduce = 1;
+long long vocab_max_size = 10000, vocab_size = 0;
+long long train_words = 0;
+real threshold = 100;
+
+unsigned long long next_random = 1;
+
+// Reads a single word from a file, assuming space + tab + EOL to be word boundaries
+void ReadWord(char *word, FILE *fin) {
+  int a = 0, ch;
+  while (!feof(fin)) {
+    ch = fgetc(fin);
+    if (ch == 13) continue;
+    if ((ch == ' ') || (ch == '\t') || (ch == '\n')) {
+      if (a > 0) {
+        if (ch == '\n') ungetc(ch, fin);
+        break;
+      }
+      if (ch == '\n') {
+        strcpy(word, (char *)"</s>");
+        return;
+      } else continue;
+    }
+    word[a] = ch;
+    a++;
+    if (a >= MAX_STRING - 1) a--;   // Truncate too long words
+  }
+  word[a] = 0;
+}
+
+// Returns hash value of a word
+int GetWordHash(char *word) {
+  unsigned long long a, hash = 1;
+  for (a = 0; a < strlen(word); a++) hash = hash * 257 + word[a];
+  hash = hash % vocab_hash_size;
+  return hash;
+}
+
+// Returns position of a word in the vocabulary; if the word is not found, returns -1
+int SearchVocab(char *word) {
+  unsigned int hash = GetWordHash(word);
+  while (1) {
+    if (vocab_hash[hash] == -1) return -1;
+    if (!strcmp(word, vocab[vocab_hash[hash]].word)) return vocab_hash[hash];
+    hash = (hash + 1) % vocab_hash_size;
+  }
+  return -1;
+}
+
+// Reads a word and returns its index in the vocabulary
+int ReadWordIndex(FILE *fin) {
+  char word[MAX_STRING];
+  ReadWord(word, fin);
+  if (feof(fin)) return -1;
+  return SearchVocab(word);
+}
+
+// Adds a word to the vocabulary
+int AddWordToVocab(char *word) {
+  unsigned int hash, length = strlen(word) + 1;
+  if (length > MAX_STRING) length = MAX_STRING;
+  vocab[vocab_size].word = (char *)calloc(length, sizeof(char));
+  strcpy(vocab[vocab_size].word, word);
+  vocab[vocab_size].cn = 0;
+  vocab_size++;
+  // Reallocate memory if needed
+  if (vocab_size + 2 >= vocab_max_size) {
+    vocab_max_size += 10000;
+    vocab=(struct vocab_word *)realloc(vocab, vocab_max_size * sizeof(struct vocab_word));
+  }
+  hash = GetWordHash(word);
+  while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+  vocab_hash[hash]=vocab_size - 1;
+  return vocab_size - 1;
+}
+
+// Used later for sorting by word counts
+int VocabCompare(const void *a, const void *b) {
+    return ((struct vocab_word *)b)->cn - ((struct vocab_word *)a)->cn;
+}
+
+// Sorts the vocabulary by frequency using word counts
+void SortVocab() {
+  int a;
+  unsigned int hash;
+  // Sort the vocabulary and keep </s> at the first position
+  qsort(&vocab[1], vocab_size - 1, sizeof(struct vocab_word), VocabCompare);
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  for (a = 0; a < vocab_size; a++) {
+    // Words occuring less than min_count times will be discarded from the vocab
+    if (vocab[a].cn < min_count) {
+      vocab_size--;
+      free(vocab[vocab_size].word);
+    } else {
+      // Hash will be re-computed, as after the sorting it is not actual
+      hash = GetWordHash(vocab[a].word);
+      while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+      vocab_hash[hash] = a;
+    }
+  }
+  vocab = (struct vocab_word *)realloc(vocab, vocab_size * sizeof(struct vocab_word));
+}
+
+// Reduces the vocabulary by removing infrequent tokens
+void ReduceVocab() {
+  int a, b = 0;
+  unsigned int hash;
+  for (a = 0; a < vocab_size; a++) if (vocab[a].cn > min_reduce) {
+    vocab[b].cn = vocab[a].cn;
+    vocab[b].word = vocab[a].word;
+    b++;
+  } else free(vocab[a].word);
+  vocab_size = b;
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  for (a = 0; a < vocab_size; a++) {
+    // Hash will be re-computed, as it is not actual
+    hash = GetWordHash(vocab[a].word);
+    while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+    vocab_hash[hash] = a;
+  }
+  fflush(stdout);
+  min_reduce++;
+}
+
+void LearnVocabFromTrainFile() {
+  char word[MAX_STRING], last_word[MAX_STRING], bigram_word[MAX_STRING * 2];
+  FILE *fin;
+  long long a, i, start = 1;
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  fin = fopen(train_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: training data file not found!\n");
+    exit(1);
+  }
+  vocab_size = 0;
+  AddWordToVocab((char *)"</s>");
+  while (1) {
+    ReadWord(word, fin);
+    if (feof(fin)) break;
+    if (!strcmp(word, "</s>")) {
+      start = 1;
+      continue;
+    } else start = 0;
+    train_words++;
+    if ((debug_mode > 1) && (train_words % 100000 == 0)) {
+      printf("Words processed: %lldK     Vocab size: %lldK  %c", train_words / 1000, vocab_size / 1000, 13);
+      fflush(stdout);
+    }
+    i = SearchVocab(word);
+    if (i == -1) {
+      a = AddWordToVocab(word);
+      vocab[a].cn = 1;
+    } else vocab[i].cn++;
+    if (start) continue;
+    sprintf(bigram_word, "%s_%s", last_word, word);
+    bigram_word[MAX_STRING - 1] = 0;
+    strcpy(last_word, word);
+    i = SearchVocab(bigram_word);
+    if (i == -1) {
+      a = AddWordToVocab(bigram_word);
+      vocab[a].cn = 1;
+    } else vocab[i].cn++;
+    if (vocab_size > vocab_hash_size * 0.7) ReduceVocab();
+  }
+  SortVocab();
+  if (debug_mode > 0) {
+    printf("\nVocab size (unigrams + bigrams): %lld\n", vocab_size);
+    printf("Words in train file: %lld\n", train_words);
+  }
+  fclose(fin);
+}
+
+void TrainModel() {
+  long long pa = 0, pb = 0, pab = 0, oov, i, li = -1, cn = 0;
+  char word[MAX_STRING], last_word[MAX_STRING], bigram_word[MAX_STRING * 2];
+  real score;
+  FILE *fo, *fin;
+  printf("Starting training using file %s\n", train_file);
+  LearnVocabFromTrainFile();
+  fin = fopen(train_file, "rb");
+  fo = fopen(output_file, "wb");
+  word[0] = 0;
+  while (1) {
+    strcpy(last_word, word);
+    ReadWord(word, fin);
+    if (feof(fin)) break;
+    if (!strcmp(word, "</s>")) {
+      fprintf(fo, "\n");
+      continue;
+    }
+    cn++;
+    if ((debug_mode > 1) && (cn % 100000 == 0)) {
+      printf("Words written: %lldK%c", cn / 1000, 13);
+      fflush(stdout);
+    }
+    oov = 0;
+    i = SearchVocab(word);
+    if (i == -1) oov = 1; else pb = vocab[i].cn;
+    if (li == -1) oov = 1;
+    li = i;
+    sprintf(bigram_word, "%s_%s", last_word, word);
+    bigram_word[MAX_STRING - 1] = 0;
+    i = SearchVocab(bigram_word);
+    if (i == -1) oov = 1; else pab = vocab[i].cn;
+    if (pa < min_count) oov = 1;
+    if (pb < min_count) oov = 1;
+    if (oov) score = 0; else score = (pab - min_count) / (real)pa / (real)pb * (real)train_words;
+    if (score > threshold) {
+      fprintf(fo, "_%s", word);
+      pb = 0;
+    } else fprintf(fo, " %s", word);
+    pa = pb;
+  }
+  fclose(fo);
+  fclose(fin);
+}
+
+int ArgPos(char *str, int argc, char **argv) {
+  int a;
+  for (a = 1; a < argc; a++) if (!strcmp(str, argv[a])) {
+    if (a == argc - 1) {
+      printf("Argument missing for %s\n", str);
+      exit(1);
+    }
+    return a;
+  }
+  return -1;
+}
+
+int main(int argc, char **argv) {
+  int i;
+  if (argc == 1) {
+    printf("WORD2PHRASE tool v0.1a\n\n");
+    printf("Options:\n");
+    printf("Parameters for training:\n");
+    printf("\t-train <file>\n");
+    printf("\t\tUse text data from <file> to train the model\n");
+    printf("\t-output <file>\n");
+    printf("\t\tUse <file> to save the resulting word vectors / word clusters / phrases\n");
+    printf("\t-min-count <int>\n");
+    printf("\t\tThis will discard words that appear less than <int> times; default is 5\n");
+    printf("\t-threshold <float>\n");
+    printf("\t\t The <float> value represents threshold for forming the phrases (higher means less phrases); default 100\n");
+    printf("\t-debug <int>\n");
+    printf("\t\tSet the debug mode (default = 2 = more info during training)\n");
+    printf("\nExamples:\n");
+    printf("./word2phrase -train text.txt -output phrases.txt -threshold 100 -debug 2\n\n");
+    return 0;
+  }
+  if ((i = ArgPos((char *)"-train", argc, argv)) > 0) strcpy(train_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-debug", argc, argv)) > 0) debug_mode = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-output", argc, argv)) > 0) strcpy(output_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-min-count", argc, argv)) > 0) min_count = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-threshold", argc, argv)) > 0) threshold = atof(argv[i + 1]);
+  vocab = (struct vocab_word *)calloc(vocab_max_size, sizeof(struct vocab_word));
+  vocab_hash = (int *)calloc(vocab_hash_size, sizeof(int));
+  TrainModel();
+  return 0;
+}
diff --git a/word2vec.c b/word2vec.c
new file mode 100644
index 0000000..67d9846
--- /dev/null
+++ b/word2vec.c
@@ -0,0 +1,1274 @@
+//  Copyright 2013 Google Inc. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <pthread.h>
+
+#define MAX_STRING 100
+#define EXP_TABLE_SIZE 1000
+#define MAX_EXP 6
+#define MAX_SENTENCE_LENGTH 1000
+#define MAX_CODE_LENGTH 40
+
+const int vocab_hash_size = 30000000;  // Maximum 30 * 0.7 = 21M words in the vocabulary
+
+typedef float real;                    // Precision of float numbers
+
+struct vocab_word {
+  long long cn;
+  int *point;
+  char *word, *code, codelen;
+};
+
+char train_file[MAX_STRING], output_file[MAX_STRING];
+char save_vocab_file[MAX_STRING], read_vocab_file[MAX_STRING];
+struct vocab_word *vocab;
+int binary = 0, type = 1, debug_mode = 2, window = 5, min_count = 5, num_threads = 12, min_reduce = 1;
+int *vocab_hash;
+long long vocab_max_size = 1000, vocab_size = 0, layer1_size = 100;
+long long train_words = 0, word_count_actual = 0, iter = 5, file_size = 0, classes = 0;
+real alpha = 0.025, starting_alpha, sample = 1e-3;
+real *syn0, *syn1, *syn1neg, *syn1nce, *expTable;
+clock_t start;
+
+real *syn1_window, *syn1neg_window, *syn1nce_window;
+int w_offset, window_layer_size;
+
+int window_hidden_size = 500; 
+real *syn_window_hidden, *syn_hidden_word, *syn_hidden_word_neg, *syn_hidden_word_nce; 
+
+int hs = 0, negative = 5;
+const int table_size = 1e8;
+int *table;
+
+//constrastive negative sampling
+char negative_classes_file[MAX_STRING];
+int *word_to_group;
+int *group_to_table; //group_size*table_size
+int class_number;
+
+//nce
+real* noise_distribution;
+int nce = 0;
+
+//param caps
+real CAP_VALUE = 50;
+int cap = 0;
+
+void capParam(real* array, int index){
+	if(array[index] > CAP_VALUE) 
+		array[index] = CAP_VALUE;
+	else if(array[index] < -CAP_VALUE)
+		array[index] = -CAP_VALUE; 
+}
+
+real hardTanh(real x){
+	if(x>=1){
+		return 1;
+	}
+	else if(x<=-1){
+		return -1;
+	}
+	else{
+		return x;
+	}
+}
+
+real dHardTanh(real x, real g){
+	if(x > 1 && g > 0){
+		return 0;
+	}
+	if(x < -1 && g < 0){
+		return 0;
+	}
+	return 1;
+}
+
+void InitUnigramTable() {
+  int a, i;
+  long long train_words_pow = 0;
+  real d1, power = 0.75;
+  table = (int *)malloc(table_size * sizeof(int));
+  for (a = 0; a < vocab_size; a++) train_words_pow += pow(vocab[a].cn, power);
+  i = 0;
+  d1 = pow(vocab[i].cn, power) / (real)train_words_pow;
+  for (a = 0; a < table_size; a++) {
+    table[a] = i;
+    if (a / (real)table_size > d1) {
+      i++;
+      d1 += pow(vocab[i].cn, power) / (real)train_words_pow;
+    }
+    if (i >= vocab_size) i = vocab_size - 1;
+  }
+  
+  noise_distribution = (real *)calloc(vocab_size, sizeof(real));
+  for (a = 0; a < vocab_size; a++) noise_distribution[a] = pow(vocab[a].cn, power)/(real)train_words_pow;
+}
+
+// Reads a single word from a file, assuming space + tab + EOL to be word boundaries
+void ReadWord(char *word, FILE *fin) {
+  int a = 0, ch;
+  while (!feof(fin)) {
+    ch = fgetc(fin);
+    if (ch == 13) continue;
+    if ((ch == ' ') || (ch == '\t') || (ch == '\n')) {
+      if (a > 0) {
+        if (ch == '\n') ungetc(ch, fin);
+        break;
+      }
+      if (ch == '\n') {
+        strcpy(word, (char *)"</s>");
+        return;
+      } else continue;
+    }
+    word[a] = ch;
+    a++;
+    if (a >= MAX_STRING - 1) a--;   // Truncate too long words
+  }
+  word[a] = 0;
+}
+
+// Returns hash value of a word
+int GetWordHash(char *word) {
+  unsigned long long a, hash = 0;
+  for (a = 0; a < strlen(word); a++) hash = hash * 257 + word[a];
+  hash = hash % vocab_hash_size;
+  return hash;
+}
+
+// Returns position of a word in the vocabulary; if the word is not found, returns -1
+int SearchVocab(char *word) {
+  unsigned int hash = GetWordHash(word);
+  while (1) {
+    if (vocab_hash[hash] == -1) return -1;
+    if (!strcmp(word, vocab[vocab_hash[hash]].word)) return vocab_hash[hash];
+    hash = (hash + 1) % vocab_hash_size;
+  }
+  return -1;
+}
+
+// Reads a word and returns its index in the vocabulary
+int ReadWordIndex(FILE *fin) {
+  char word[MAX_STRING];
+  ReadWord(word, fin);
+  if (feof(fin)) return -1;
+  return SearchVocab(word);
+}
+
+// Adds a word to the vocabulary
+int AddWordToVocab(char *word) {
+  unsigned int hash, length = strlen(word) + 1;
+  if (length > MAX_STRING) length = MAX_STRING;
+  vocab[vocab_size].word = (char *)calloc(length, sizeof(char));
+  strcpy(vocab[vocab_size].word, word);
+  vocab[vocab_size].cn = 0;
+  vocab_size++;
+  // Reallocate memory if needed
+  if (vocab_size + 2 >= vocab_max_size) {
+    vocab_max_size += 1000;
+    vocab = (struct vocab_word *)realloc(vocab, vocab_max_size * sizeof(struct vocab_word));
+  }
+  hash = GetWordHash(word);
+  while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+  vocab_hash[hash] = vocab_size - 1;
+  return vocab_size - 1;
+}
+
+// Used later for sorting by word counts
+int VocabCompare(const void *a, const void *b) {
+    return ((struct vocab_word *)b)->cn - ((struct vocab_word *)a)->cn;
+}
+
+// Sorts the vocabulary by frequency using word counts
+void SortVocab() {
+  int a, size;
+  unsigned int hash;
+  // Sort the vocabulary and keep </s> at the first position
+  qsort(&vocab[1], vocab_size - 1, sizeof(struct vocab_word), VocabCompare);
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  size = vocab_size;
+  train_words = 0;
+  for (a = 0; a < size; a++) {
+    // Words occuring less than min_count times will be discarded from the vocab
+    if ((vocab[a].cn < min_count) && (a != 0)) {
+      vocab_size--;
+      free(vocab[a].word);
+    } else {
+      // Hash will be re-computed, as after the sorting it is not actual
+      hash=GetWordHash(vocab[a].word);
+      while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+      vocab_hash[hash] = a;
+      train_words += vocab[a].cn;
+    }
+  }
+  vocab = (struct vocab_word *)realloc(vocab, (vocab_size + 1) * sizeof(struct vocab_word));
+  // Allocate memory for the binary tree construction
+  for (a = 0; a < vocab_size; a++) {
+    vocab[a].code = (char *)calloc(MAX_CODE_LENGTH, sizeof(char));
+    vocab[a].point = (int *)calloc(MAX_CODE_LENGTH, sizeof(int));
+  }
+}
+
+// Reduces the vocabulary by removing infrequent tokens
+void ReduceVocab() {
+  int a, b = 0;
+  unsigned int hash;
+  for (a = 0; a < vocab_size; a++) if (vocab[a].cn > min_reduce) {
+    vocab[b].cn = vocab[a].cn;
+    vocab[b].word = vocab[a].word;
+    b++;
+  } else free(vocab[a].word);
+  vocab_size = b;
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  for (a = 0; a < vocab_size; a++) {
+    // Hash will be re-computed, as it is not actual
+    hash = GetWordHash(vocab[a].word);
+    while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+    vocab_hash[hash] = a;
+  }
+  fflush(stdout);
+  min_reduce++;
+}
+
+// Create binary Huffman tree using the word counts
+// Frequent words will have short uniqe binary codes
+void CreateBinaryTree() {
+  long long a, b, i, min1i, min2i, pos1, pos2, point[MAX_CODE_LENGTH];
+  char code[MAX_CODE_LENGTH];
+  long long *count = (long long *)calloc(vocab_size * 2 + 1, sizeof(long long));
+  long long *binary = (long long *)calloc(vocab_size * 2 + 1, sizeof(long long));
+  long long *parent_node = (long long *)calloc(vocab_size * 2 + 1, sizeof(long long));
+  for (a = 0; a < vocab_size; a++) count[a] = vocab[a].cn;
+  for (a = vocab_size; a < vocab_size * 2; a++) count[a] = 1e15;
+  pos1 = vocab_size - 1;
+  pos2 = vocab_size;
+  // Following algorithm constructs the Huffman tree by adding one node at a time
+  for (a = 0; a < vocab_size - 1; a++) {
+    // First, find two smallest nodes 'min1, min2'
+    if (pos1 >= 0) {
+      if (count[pos1] < count[pos2]) {
+        min1i = pos1;
+        pos1--;
+      } else {
+        min1i = pos2;
+        pos2++;
+      }
+    } else {
+      min1i = pos2;
+      pos2++;
+    }
+    if (pos1 >= 0) {
+      if (count[pos1] < count[pos2]) {
+        min2i = pos1;
+        pos1--;
+      } else {
+        min2i = pos2;
+        pos2++;
+      }
+    } else {
+      min2i = pos2;
+      pos2++;
+    }
+    count[vocab_size + a] = count[min1i] + count[min2i];
+    parent_node[min1i] = vocab_size + a;
+    parent_node[min2i] = vocab_size + a;
+    binary[min2i] = 1;
+  }
+  // Now assign binary code to each vocabulary word
+  for (a = 0; a < vocab_size; a++) {
+    b = a;
+    i = 0;
+    while (1) {
+      code[i] = binary[b];
+      point[i] = b;
+      i++;
+      b = parent_node[b];
+      if (b == vocab_size * 2 - 2) break;
+    }
+    vocab[a].codelen = i;
+    vocab[a].point[0] = vocab_size - 2;
+    for (b = 0; b < i; b++) {
+      vocab[a].code[i - b - 1] = code[b];
+      vocab[a].point[i - b] = point[b] - vocab_size;
+    }
+  }
+  free(count);
+  free(binary);
+  free(parent_node);
+}
+
+void LearnVocabFromTrainFile() {
+  char word[MAX_STRING];
+  FILE *fin;
+  long long a, i;
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  fin = fopen(train_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: training data file not found!\n");
+    exit(1);
+  }
+  vocab_size = 0;
+  AddWordToVocab((char *)"</s>");
+  while (1) {
+    ReadWord(word, fin);
+    if (feof(fin)) break;
+    train_words++;
+    if ((debug_mode > 1) && (train_words % 100000 == 0)) {
+      printf("%lldK%c", train_words / 1000, 13);
+      fflush(stdout);
+    }
+    i = SearchVocab(word);
+    if (i == -1) {
+      a = AddWordToVocab(word);
+      vocab[a].cn = 1;
+    } else vocab[i].cn++;
+    if (vocab_size > vocab_hash_size * 0.7) ReduceVocab();
+  }
+  SortVocab();
+  if (debug_mode > 0) {
+    printf("Vocab size: %lld\n", vocab_size);
+    printf("Words in train file: %lld\n", train_words);
+  }
+  file_size = ftell(fin);
+  fclose(fin);
+}
+
+void SaveVocab() {
+  long long i;
+  FILE *fo = fopen(save_vocab_file, "wb");
+  for (i = 0; i < vocab_size; i++) fprintf(fo, "%s %lld\n", vocab[i].word, vocab[i].cn);
+  fclose(fo);
+}
+
+void ReadVocab() {
+  long long a, i = 0;
+  char c;
+  char word[MAX_STRING];
+  FILE *fin = fopen(read_vocab_file, "rb");
+  if (fin == NULL) {
+    printf("Vocabulary file not found\n");
+    exit(1);
+  }
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  vocab_size = 0;
+  while (1) {
+    ReadWord(word, fin);
+    if (feof(fin)) break;
+    a = AddWordToVocab(word);
+    fscanf(fin, "%lld%c", &vocab[a].cn, &c);
+    i++;
+  }
+  SortVocab();
+  if (debug_mode > 0) {
+    printf("Vocab size: %lld\n", vocab_size);
+    printf("Words in train file: %lld\n", train_words);
+  }
+  fin = fopen(train_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: training data file not found!\n");
+    exit(1);
+  }
+  fseek(fin, 0, SEEK_END);
+  file_size = ftell(fin);
+  fclose(fin);
+}
+
+void InitClassUnigramTable() {
+  long long a,c;
+  printf("loading class unigrams \n");
+  FILE *fin = fopen(negative_classes_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: class file not found!\n");
+    exit(1);
+  }
+  word_to_group = (int *)malloc(vocab_size * sizeof(int));
+  for(a = 0; a < vocab_size; a++) word_to_group[a] = -1;
+  char class[MAX_STRING];
+  char prev_class[MAX_STRING];
+  prev_class[0] = 0;
+  char word[MAX_STRING];
+  class_number = -1;
+  while (1) {
+    if (feof(fin)) break;
+    ReadWord(class, fin);
+    ReadWord(word, fin);
+    int word_index = SearchVocab(word);
+    if (word_index != -1){
+       if(strcmp(class, prev_class) != 0){
+	    class_number++;
+	    strcpy(prev_class, class);
+       }
+       word_to_group[word_index] = class_number;
+    }
+    ReadWord(word, fin);
+  }
+  class_number++;
+  fclose(fin);
+  
+  group_to_table = (int *)malloc(table_size * class_number * sizeof(int)); 
+  long long train_words_pow = 0;
+  real d1, power = 0.75;
+  
+  for(c = 0; c < class_number; c++){
+     long long offset = c * table_size;
+     train_words_pow = 0;
+     for (a = 0; a < vocab_size; a++) if(word_to_group[a] == c) train_words_pow += pow(vocab[a].cn, power);
+     int i = 0;
+     while(word_to_group[i]!=c && i < vocab_size) i++;
+     d1 = pow(vocab[i].cn, power) / (real)train_words_pow;
+     for (a = 0; a < table_size; a++) {
+	//printf("index %lld , word %d\n", a, i);
+	group_to_table[offset + a] = i;
+        if (a / (real)table_size > d1) {
+	   i++;
+           while(word_to_group[i]!=c && i < vocab_size) i++;
+	   d1 += pow(vocab[i].cn, power) / (real)train_words_pow;
+        }
+        if (i >= vocab_size) while(word_to_group[i]!=c && i >= 0) i--;
+     }
+  }
+}
+
+void InitNet() {
+  long long a, b;
+  unsigned long long next_random = 1;
+  window_layer_size = layer1_size*window*2;
+  a = posix_memalign((void **)&syn0, 128, (long long)vocab_size * layer1_size * sizeof(real));
+  if (syn0 == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  
+  if (hs) {
+    a = posix_memalign((void **)&syn1, 128, (long long)vocab_size * layer1_size * sizeof(real));
+    if (syn1 == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn1_window, 128, (long long)vocab_size * window_layer_size * sizeof(real));
+    if (syn1_window == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn_hidden_word, 128, (long long)vocab_size * window_hidden_size * sizeof(real));
+    if (syn_hidden_word == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++)
+     syn1[a * layer1_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_layer_size; b++)
+     syn1_window[a * window_layer_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_hidden_size; b++)
+     syn_hidden_word[a * window_hidden_size + b] = 0;
+  }
+  if (negative>0) {
+    a = posix_memalign((void **)&syn1neg, 128, (long long)vocab_size * layer1_size * sizeof(real));
+    if (syn1neg == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn1neg_window, 128, (long long)vocab_size * window_layer_size * sizeof(real));
+    if (syn1neg_window == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn_hidden_word_neg, 128, (long long)vocab_size * window_hidden_size * sizeof(real));
+    if (syn_hidden_word_neg == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++)
+     syn1neg[a * layer1_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_layer_size; b++)
+     syn1neg_window[a * window_layer_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_hidden_size; b++)
+     syn_hidden_word_neg[a * window_hidden_size + b] = 0;
+  }
+  if (nce>0) {
+    a = posix_memalign((void **)&syn1nce, 128, (long long)vocab_size * layer1_size * sizeof(real));
+    if (syn1nce == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn1nce_window, 128, (long long)vocab_size * window_layer_size * sizeof(real));
+    if (syn1nce_window == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn_hidden_word_nce, 128, (long long)vocab_size * window_hidden_size * sizeof(real));
+    if (syn_hidden_word_nce == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++)
+     syn1nce[a * layer1_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_layer_size; b++)
+     syn1nce_window[a * window_layer_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_hidden_size; b++)
+     syn_hidden_word_nce[a * window_hidden_size + b] = 0;
+  }
+  for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++) {
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    syn0[a * layer1_size + b] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / layer1_size;
+  }
+
+  a = posix_memalign((void **)&syn_window_hidden, 128, window_hidden_size * window_layer_size * sizeof(real));
+  if (syn_window_hidden == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  for (a = 0; a < window_hidden_size * window_layer_size; a++){
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    syn_window_hidden[a] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / (window_hidden_size*window_layer_size);
+  }
+
+  CreateBinaryTree();
+}
+
+void *TrainModelThread(void *id) {
+  long long a, b, d, cw, word, last_word, sentence_length = 0, sentence_position = 0;
+  long long word_count = 0, last_word_count = 0, sen[MAX_SENTENCE_LENGTH + 1];
+  long long l1, l2, c, target, label, local_iter = iter;
+  unsigned long long next_random = (long long)id;
+  real f, g;
+  clock_t now;
+  int input_len_1 = layer1_size;
+  int window_offset = -1;
+  if(type == 2 || type == 4){
+     input_len_1=window_layer_size;
+  }
+  real *neu1 = (real *)calloc(input_len_1, sizeof(real));
+  real *neu1e = (real *)calloc(input_len_1, sizeof(real));
+
+  int input_len_2 = 0;
+  if(type == 4){
+     input_len_2 = window_hidden_size;
+  }
+  real *neu2 = (real *)calloc(input_len_2, sizeof(real));
+  real *neu2e = (real *)calloc(input_len_2, sizeof(real));
+
+  FILE *fi = fopen(train_file, "rb");
+  fseek(fi, file_size / (long long)num_threads * (long long)id, SEEK_SET);
+  while (1) {
+    if (word_count - last_word_count > 10000) {
+      word_count_actual += word_count - last_word_count;
+      last_word_count = word_count;
+      if ((debug_mode > 1)) {
+        now=clock();
+        printf("%cAlpha: %f  Progress: %.2f%%  Words/thread/sec: %.2fk  ", 13, alpha,
+         word_count_actual / (real)(iter * train_words + 1) * 100,
+         word_count_actual / ((real)(now - start + 1) / (real)CLOCKS_PER_SEC * 1000));
+        fflush(stdout);
+      }
+      alpha = starting_alpha * (1 - word_count_actual / (real)(iter * train_words + 1));
+      if (alpha < starting_alpha * 0.0001) alpha = starting_alpha * 0.0001;
+    }
+    if (sentence_length == 0) {
+      while (1) {
+        word = ReadWordIndex(fi);
+        if (feof(fi)) break;
+        if (word == -1) continue;
+        word_count++;
+        if (word == 0) break;
+        // The subsampling randomly discards frequent words while keeping the ranking same
+        if (sample > 0) {
+          real ran = (sqrt(vocab[word].cn / (sample * train_words)) + 1) * (sample * train_words) / vocab[word].cn;
+          next_random = next_random * (unsigned long long)25214903917 + 11;
+          if (ran < (next_random & 0xFFFF) / (real)65536) continue;
+        }
+        sen[sentence_length] = word;
+        sentence_length++;
+        if (sentence_length >= MAX_SENTENCE_LENGTH) break;
+      }
+      sentence_position = 0;
+    }
+    if (feof(fi) || (word_count > train_words / num_threads)) {
+      word_count_actual += word_count - last_word_count;
+      local_iter--;
+      if (local_iter == 0) break;
+      word_count = 0;
+      last_word_count = 0;
+      sentence_length = 0;
+      fseek(fi, file_size / (long long)num_threads * (long long)id, SEEK_SET);
+      continue;
+    }
+    word = sen[sentence_position];
+    if (word == -1) continue;
+    for (c = 0; c < input_len_1; c++) neu1[c] = 0;
+    for (c = 0; c < input_len_1; c++) neu1e[c] = 0;
+    for (c = 0; c < input_len_2; c++) neu2[c] = 0;
+    for (c = 0; c < input_len_2; c++) neu2e[c] = 0;
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    b = next_random % window;
+    if (type == 0) {  //train the cbow architecture
+      // in -> hidden
+      cw = 0;
+      for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        for (c = 0; c < layer1_size; c++) neu1[c] += syn0[c + last_word * layer1_size];
+        cw++;
+      }
+      if (cw) {
+        for (c = 0; c < layer1_size; c++) neu1[c] /= cw;
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * layer1_size;
+          // Propagate hidden -> output
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < layer1_size; c++) syn1[c + l2] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1, c + l2);
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+	    if(word_to_group != NULL && word_to_group[word] != -1){
+		target = word;
+		while(target == word) {
+			target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+            		next_random = next_random * (unsigned long long)25214903917 + 11;
+		}
+		//printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+	    }
+	    else{
+            	target = table[(next_random >> 16) % table_size];
+	    }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1neg[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha;
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1neg[c + l2] += g * neu1[c];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1neg, c + l2);
+        }
+        // Noise Contrastive Estimation
+        if (nce > 0) for (d = 0; d < nce + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+	    if(word_to_group != NULL && word_to_group[word] != -1){
+		target = word;
+		while(target == word) {
+			target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+            		next_random = next_random * (unsigned long long)25214903917 + 11;
+		}
+	    }
+	    else{
+            	target = table[(next_random >> 16) % table_size];
+	    }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+	  
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1nce[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else {
+                f = exp(f);
+                g = (label - f/(noise_distribution[target]*nce + f)) * alpha;
+          }
+	  for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1nce[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1nce[c + l2] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1nce,c + l2);
+        }
+        // hidden -> in
+        for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {
+          c = sentence_position - window + a;
+          if (c < 0) continue;
+          if (c >= sentence_length) continue;
+          last_word = sen[c];
+          if (last_word == -1) continue;
+          for (c = 0; c < layer1_size; c++) syn0[c + last_word * layer1_size] += neu1e[c];
+        }
+      }
+    } else if(type==1) {  //train skip-gram
+      for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        l1 = last_word * layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1e[c] = 0;
+        // HIERARCHICAL SOFTMAX
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * layer1_size;
+          // Propagate hidden -> output
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < layer1_size; c++) syn1[c + l2] += g * syn0[c + l1];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1, c + l2);
+        }
+        // NEGATIVE SAMPLING
+	if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	    next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1neg[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha;
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1neg[c + l2] += g * syn0[c + l1];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1neg, c + l2);
+        }
+	//Noise Contrastive Estimation
+        if (nce > 0) for (d = 0; d < nce + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	    next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1nce[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else {
+                f = exp(f);
+                g = (label - f/(noise_distribution[target]*nce + f)) * alpha;
+          }
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1nce[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1nce[c + l2] += g * syn0[c + l1];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1nce, c + l2);
+        }
+        // Learn weights input -> hidden
+        for (c = 0; c < layer1_size; c++) syn0[c + l1] += neu1e[c];
+      }
+    }
+    else if(type == 2){ //train the cwindow architecture
+      // in -> hidden
+      cw = 0;
+      for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        window_offset = a*layer1_size;
+        if (a > window) window_offset-=layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1[c+window_offset] += syn0[c + last_word * layer1_size];
+        cw++;
+      }
+      if (cw) {
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * window_layer_size;
+          // Propagate hidden -> output
+          for (c = 0; c < window_layer_size; c++) f += neu1[c] * syn1_window[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < window_layer_size; c++) neu1e[c] += g * syn1_window[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < window_layer_size; c++) syn1_window[c + l2] += g * neu1[c];
+	  if (cap == 1) for (c = 0; c < window_layer_size; c++) capParam(syn1_window, c + l2);
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < window_layer_size; c++) f += neu1[c] * syn1neg_window[c + l2];
+	  if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha;
+          for (c = 0; c < window_layer_size; c++) neu1e[c] += g * syn1neg_window[c + l2];
+          for (c = 0; c < window_layer_size; c++) syn1neg_window[c + l2] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < window_layer_size; c++) capParam(syn1neg_window, c + l2);
+        }
+	// Noise Contrastive Estimation
+        if (nce > 0) for (d = 0; d < nce + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < window_layer_size; c++) f += neu1[c] * syn1nce_window[c + l2];
+	  if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else {
+                f = exp(f);
+                g = (label - f/(noise_distribution[target]*nce + f)) * alpha;
+          }
+          for (c = 0; c < window_layer_size; c++) neu1e[c] += g * syn1nce_window[c + l2];
+          for (c = 0; c < window_layer_size; c++) syn1nce_window[c + l2] += g * neu1[c];
+	  if(cap == 1) for (c = 0; c < window_layer_size; c++) capParam(syn1nce_window, c + l2);
+        }
+        // hidden -> in
+        for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+          c = sentence_position - window + a;
+          if (c < 0) continue;
+          if (c >= sentence_length) continue;
+          last_word = sen[c];
+          if (last_word == -1) continue;
+	  window_offset = a * layer1_size;
+	  if(a > window) window_offset -= layer1_size;
+          for (c = 0; c < layer1_size; c++) syn0[c + last_word * layer1_size] += neu1e[c + window_offset];
+        }
+      }
+    }
+    else if (type == 3){  //train structured skip-gram
+      for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        l1 = last_word * layer1_size;
+	window_offset = a * layer1_size;
+	if(a > window) window_offset -= layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1e[c] = 0;
+        // HIERARCHICAL SOFTMAX
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * window_layer_size;
+          // Propagate hidden -> output
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1_window[c + l2 + window_offset];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1_window[c + l2 + window_offset];
+          // Learn weights hidden -> output
+          for (c = 0; c < layer1_size; c++) syn1[c + l2 + window_offset] += g * syn0[c + l1];
+	  if(cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1, c + l2 + window_offset);
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	     next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1neg_window[c + l2 + window_offset];
+	  if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha;
+	  for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg_window[c + l2 + window_offset];
+          for (c = 0; c < layer1_size; c++) syn1neg_window[c + l2 + window_offset] += g * syn0[c + l1]; 
+	  if(cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1neg_window, c + l2 + window_offset);
+        }
+	// Noise Constrastive Estimation
+        if (nce > 0) for (d = 0; d < nce + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+             next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1nce_window[c + l2 + window_offset];
+	  if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else {
+                f = exp(f);
+                g = (label - f/(noise_distribution[target]*nce + f)) * alpha;
+          }
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1nce_window[c + l2 + window_offset];
+          for (c = 0; c < layer1_size; c++) syn1nce_window[c + l2 + window_offset] += g * syn0[c + l1];
+	  if (cap == 1) for (c = 0; c < layer1_size; c++) capParam(syn1nce_window, c + l2 + window_offset);
+        }
+        // Learn weights input -> hidden
+        for (c = 0; c < layer1_size; c++) {syn0[c + l1] += neu1e[c]; if(syn0[c + l1] > 50) syn0[c + l1] = 50; if(syn0[c + l1] < -50) syn0[c + l1] = -50;}
+      }
+    }
+    else if(type == 4){ //training senna
+	// in -> hidden
+      cw = 0;
+      for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        window_offset = a*layer1_size;
+        if (a > window) window_offset-=layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1[c+window_offset] += syn0[c + last_word * layer1_size];
+        cw++;
+      }
+      if (cw) {
+		for (a = 0; a < window_hidden_size; a++){
+          c = a*window_layer_size;
+          for(b = 0; b < window_layer_size; b++){
+             neu2[a] += syn_window_hidden[c + b] * neu1[b];
+          }
+        }
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * window_hidden_size;
+          // Propagate hidden -> output
+          for (c = 0; c < window_hidden_size; c++) f += hardTanh(neu2[c]) * syn_hidden_word[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < window_hidden_size; c++) neu2e[c] += dHardTanh(neu2[c],g) * g * syn_hidden_word[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < window_hidden_size; c++) syn_hidden_word[c + l2] += dHardTanh(neu2[c],g) * g * neu2[c];
+        }
+      // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	    next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_hidden_size;
+          f = 0;
+          for (c = 0; c < window_hidden_size; c++) f += hardTanh(neu2[c]) * syn_hidden_word_neg[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha / negative;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha / negative;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha / negative;
+          for (c = 0; c < window_hidden_size; c++) neu2e[c] += dHardTanh(neu2[c],g) * g * syn_hidden_word_neg[c + l2];
+          for (c = 0; c < window_hidden_size; c++) syn_hidden_word_neg[c + l2] += dHardTanh(neu2[c],g) * g * neu2[c];
+        }
+        for (a = 0; a < window_hidden_size; a++)
+          for(b = 0; b < window_layer_size; b++)
+	     neu1e[b] += neu2e[a] * syn_window_hidden[a*window_layer_size + b];
+	for (a = 0; a < window_hidden_size; a++)
+          for(b = 0; b < window_layer_size; b++)
+	     syn_window_hidden[a*window_layer_size + b] += neu2e[a] * neu1[b];
+        // hidden -> in
+        for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+          c = sentence_position - window + a;
+          if (c < 0) continue;
+          if (c >= sentence_length) continue;
+          last_word = sen[c];
+          if (last_word == -1) continue;
+          window_offset = a * layer1_size;
+          if(a > window) window_offset -= layer1_size;
+          for (c = 0; c < layer1_size; c++) syn0[c + last_word * layer1_size] += neu1e[c + window_offset];
+        }
+      }
+    }
+    else{
+	printf("unknown type %i", type);
+	exit(0);
+    }
+    sentence_position++;
+    if (sentence_position >= sentence_length) {
+      sentence_length = 0;
+      continue;
+    }
+  }
+  fclose(fi);
+  free(neu1);
+  free(neu1e);
+  pthread_exit(NULL);
+}
+
+void TrainModel() {
+  long a, b, c, d;
+  FILE *fo;
+  pthread_t *pt = (pthread_t *)malloc(num_threads * sizeof(pthread_t));
+  printf("Starting training using file %s\n", train_file);
+  starting_alpha = alpha;
+  if (read_vocab_file[0] != 0) ReadVocab(); else LearnVocabFromTrainFile();
+  if (save_vocab_file[0] != 0) SaveVocab();
+  if (output_file[0] == 0) return;
+  InitNet();
+  if (negative > 0 || nce > 0) InitUnigramTable();
+  if (negative_classes_file[0] != 0) InitClassUnigramTable();
+  start = clock();
+  for (a = 0; a < num_threads; a++) pthread_create(&pt[a], NULL, TrainModelThread, (void *)a);
+  for (a = 0; a < num_threads; a++) pthread_join(pt[a], NULL);
+  fo = fopen(output_file, "wb");
+  if (classes == 0) {
+    // Save the word vectors
+    fprintf(fo, "%lld %lld\n", vocab_size, layer1_size);
+    for (a = 0; a < vocab_size; a++) {
+      fprintf(fo, "%s ", vocab[a].word);
+      if (binary) for (b = 0; b < layer1_size; b++) fwrite(&syn0[a * layer1_size + b], sizeof(real), 1, fo);
+      else for (b = 0; b < layer1_size; b++) fprintf(fo, "%lf ", syn0[a * layer1_size + b]);
+      fprintf(fo, "\n");
+    }
+  } else {
+    // Run K-means on the word vectors
+    int clcn = classes, iter = 10, closeid;
+    int *centcn = (int *)malloc(classes * sizeof(int));
+    int *cl = (int *)calloc(vocab_size, sizeof(int));
+    real closev, x;
+    real *cent = (real *)calloc(classes * layer1_size, sizeof(real));
+    for (a = 0; a < vocab_size; a++) cl[a] = a % clcn;
+    for (a = 0; a < iter; a++) {
+      for (b = 0; b < clcn * layer1_size; b++) cent[b] = 0;
+      for (b = 0; b < clcn; b++) centcn[b] = 1;
+      for (c = 0; c < vocab_size; c++) {
+        for (d = 0; d < layer1_size; d++) cent[layer1_size * cl[c] + d] += syn0[c * layer1_size + d];
+        centcn[cl[c]]++;
+      }
+      for (b = 0; b < clcn; b++) {
+        closev = 0;
+        for (c = 0; c < layer1_size; c++) {
+          cent[layer1_size * b + c] /= centcn[b];
+          closev += cent[layer1_size * b + c] * cent[layer1_size * b + c];
+        }
+        closev = sqrt(closev);
+        for (c = 0; c < layer1_size; c++) cent[layer1_size * b + c] /= closev;
+      }
+      for (c = 0; c < vocab_size; c++) {
+        closev = -10;
+        closeid = 0;
+        for (d = 0; d < clcn; d++) {
+          x = 0;
+          for (b = 0; b < layer1_size; b++) x += cent[layer1_size * d + b] * syn0[c * layer1_size + b];
+          if (x > closev) {
+            closev = x;
+            closeid = d;
+          }
+        }
+        cl[c] = closeid;
+      }
+    }
+    // Save the K-means classes
+    for (a = 0; a < vocab_size; a++) fprintf(fo, "%s %d\n", vocab[a].word, cl[a]);
+    free(centcn);
+    free(cent);
+    free(cl);
+  }
+  fclose(fo);
+}
+
+int ArgPos(char *str, int argc, char **argv) {
+  int a;
+  for (a = 1; a < argc; a++) if (!strcmp(str, argv[a])) {
+    if (a == argc - 1) {
+      printf("Argument missing for %s\n", str);
+      exit(1);
+    }
+    return a;
+  }
+  return -1;
+}
+
+int main(int argc, char **argv) {
+  int i;
+  if (argc == 1) {
+    printf("WORD VECTOR estimation toolkit v 0.1c\n\n");
+    printf("Options:\n");
+    printf("Parameters for training:\n");
+    printf("\t-train <file>\n");
+    printf("\t\tUse text data from <file> to train the model\n");
+    printf("\t-output <file>\n");
+    printf("\t\tUse <file> to save the resulting word vectors / word clusters\n");
+    printf("\t-size <int>\n");
+    printf("\t\tSet size of word vectors; default is 100\n");
+    printf("\t-window <int>\n");
+    printf("\t\tSet max skip length between words; default is 5\n");
+    printf("\t-sample <float>\n");
+    printf("\t\tSet threshold for occurrence of words. Those that appear with higher frequency in the training data\n");
+    printf("\t\twill be randomly down-sampled; default is 1e-3, useful range is (0, 1e-5)\n");
+    printf("\t-hs <int>\n");
+    printf("\t\tUse Hierarchical Softmax; default is 0 (not used)\n");
+    printf("\t-negative <int>\n");
+    printf("\t\tNumber of negative examples; default is 5, common values are 3 - 10 (0 = not used)\n");
+    printf("\t-negative-classes <file>\n");
+    printf("\t\tNegative classes to sample from\n");
+    printf("\t-nce <int>\n");
+    printf("\t\tNumber of negative examples for nce; default is 0, common values are 3 - 10 (0 = not used)\n");
+    printf("\t-threads <int>\n");
+    printf("\t\tUse <int> threads (default 12)\n");
+    printf("\t-iter <int>\n");
+    printf("\t\tRun more training iterations (default 5)\n");
+    printf("\t-min-count <int>\n");
+    printf("\t\tThis will discard words that appear less than <int> times; default is 5\n");
+    printf("\t-alpha <float>\n");
+    printf("\t\tSet the starting learning rate; default is 0.025 for skip-gram and 0.05 for CBOW\n");
+    printf("\t-classes <int>\n");
+    printf("\t\tOutput word classes rather than word vectors; default number of classes is 0 (vectors are written)\n");
+    printf("\t-debug <int>\n");
+    printf("\t\tSet the debug mode (default = 2 = more info during training)\n");
+    printf("\t-binary <int>\n");
+    printf("\t\tSave the resulting vectors in binary moded; default is 0 (off)\n");
+    printf("\t-save-vocab <file>\n");
+    printf("\t\tThe vocabulary will be saved to <file>\n");
+    printf("\t-read-vocab <file>\n");
+    printf("\t\tThe vocabulary will be read from <file>, not constructed from the training data\n");
+    printf("\t-type <int>\n");
+    printf("\t\tType of embeddings (0 for cbow, 1 for skipngram, 2 for cwindow, 3 for structured skipngram, 4 for senna type)\n");
+    printf("\t-cap <int>\n");
+    printf("\t\tlimit the parameter values to the range [-50, 50]; default is 0 (off)\n");
+    printf("\nExamples:\n");
+    printf("./word2vec -train data.txt -output vec.txt -size 200 -window 5 -sample 1e-4 -negative 5 -hs 0 -binary 0 -type 1 -iter 3\n\n");
+    return 0;
+  }
+  output_file[0] = 0;
+  save_vocab_file[0] = 0;
+  read_vocab_file[0] = 0;
+  negative_classes_file[0] = 0;
+  if ((i = ArgPos((char *)"-size", argc, argv)) > 0) layer1_size = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-train", argc, argv)) > 0) strcpy(train_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-save-vocab", argc, argv)) > 0) strcpy(save_vocab_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-read-vocab", argc, argv)) > 0) strcpy(read_vocab_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-debug", argc, argv)) > 0) debug_mode = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-binary", argc, argv)) > 0) binary = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-type", argc, argv)) > 0) type = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-output", argc, argv)) > 0) strcpy(output_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-window", argc, argv)) > 0) window = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-sample", argc, argv)) > 0) sample = atof(argv[i + 1]);
+  if ((i = ArgPos((char *)"-hs", argc, argv)) > 0) hs = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-negative", argc, argv)) > 0) negative = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-negative-classes", argc, argv)) > 0) strcpy(negative_classes_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-nce", argc, argv)) > 0) nce = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-threads", argc, argv)) > 0) num_threads = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-iter", argc, argv)) > 0) iter = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-min-count", argc, argv)) > 0) min_count = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-classes", argc, argv)) > 0) classes = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-cap", argc, argv)) > 0) cap = atoi(argv[i + 1]);
+  if (type==0 || type==2 || type==4) alpha = 0.05;
+  if ((i = ArgPos((char *)"-alpha", argc, argv)) > 0) alpha = atof(argv[i + 1]);
+  vocab = (struct vocab_word *)calloc(vocab_max_size, sizeof(struct vocab_word));
+  vocab_hash = (int *)calloc(vocab_hash_size, sizeof(int));
+  expTable = (real *)malloc((EXP_TABLE_SIZE + 1) * sizeof(real));
+  for (i = 0; i < EXP_TABLE_SIZE; i++) {
+    expTable[i] = exp((i / (real)EXP_TABLE_SIZE * 2 - 1) * MAX_EXP); // Precompute the exp() table
+    expTable[i] = expTable[i] / (expTable[i] + 1);                   // Precompute f(x) = x / (x + 1)
+  }
+  TrainModel();
+  return 0;
+}
+
diff --git a/word2vecExt.c b/word2vecExt.c
new file mode 100644
index 0000000..88d7ef0
--- /dev/null
+++ b/word2vecExt.c
@@ -0,0 +1,1830 @@
+//  Copyright 2013 Google Inc. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <pthread.h>
+
+#define MAX_STRING 100
+#define EXP_TABLE_SIZE 1000
+#define MAX_EXP 6
+#define MAX_SENTENCE_LENGTH 1000
+#define MAX_CODE_LENGTH 40
+
+const int vocab_hash_size = 30000000; // Maximum 30 * 0.7 = 21M words in the vocabulary
+
+typedef float real;                    // Precision of float numbers
+
+struct vocab_word {
+	long long cn;
+	int *point;
+	char *word, *code, codelen;
+};
+
+char train_file[MAX_STRING], output_file[MAX_STRING];
+char save_vocab_file[MAX_STRING], read_vocab_file[MAX_STRING];
+char save_net_file[MAX_STRING], read_net_file[MAX_STRING];
+struct vocab_word *vocab;
+int binary = 0, type = 1, debug_mode = 2, window = 5, min_count = 5,
+		num_threads = 12, min_reduce = 1;
+int *vocab_hash;
+long long vocab_max_size = 1000, vocab_size = 0, layer1_size = 100;
+long long train_words = 0, word_count_actual = 0, iter = 5, file_size = 0,
+		classes = 0;
+real alpha = 0.025, starting_alpha, sample = 1e-3;
+real *syn0, *syn1, *syn1neg, *syn1nce, *expTable;
+clock_t start;
+
+real *syn1_window, *syn1neg_window, *syn1nce_window;
+int w_offset, window_layer_size;
+
+int window_hidden_size = 500;
+real *syn_window_hidden, *syn_hidden_word, *syn_hidden_word_neg,
+		*syn_hidden_word_nce;
+
+int hs = 0, negative = 5;
+const int table_size = 1e8;
+int *table;
+
+//constrastive negative sampling
+char negative_classes_file[MAX_STRING];
+int *word_to_group;
+int *group_to_table; //group_size*table_size
+int class_number;
+
+//nce
+real* noise_distribution;
+int nce = 0;
+
+//param caps
+real CAP_VALUE = 50;
+int cap = 0;
+
+void capParam(real* array, int index) {
+	if (array[index] > CAP_VALUE)
+		array[index] = CAP_VALUE;
+	else if (array[index] < -CAP_VALUE)
+		array[index] = -CAP_VALUE;
+}
+
+real hardTanh(real x) {
+	if (x >= 1) {
+		return 1;
+	} else if (x <= -1) {
+		return -1;
+	} else {
+		return x;
+	}
+}
+
+real dHardTanh(real x, real g) {
+	if (x > 1 && g > 0) {
+		return 0;
+	}
+	if (x < -1 && g < 0) {
+		return 0;
+	}
+	return 1;
+}
+
+void InitUnigramTable() {
+	int a, i;
+	long long train_words_pow = 0;
+	real d1, power = 0.75;
+	table = (int *) malloc(table_size * sizeof(int));
+	for (a = 0; a < vocab_size; a++)
+		train_words_pow += pow(vocab[a].cn, power);
+	i = 0;
+	d1 = pow(vocab[i].cn, power) / (real) train_words_pow;
+	for (a = 0; a < table_size; a++) {
+		table[a] = i;
+		if (a / (real) table_size > d1) {
+			i++;
+			d1 += pow(vocab[i].cn, power) / (real) train_words_pow;
+		}
+		if (i >= vocab_size)
+			i = vocab_size - 1;
+	}
+
+	noise_distribution = (real *) calloc(vocab_size, sizeof(real));
+	for (a = 0; a < vocab_size; a++)
+		noise_distribution[a] = pow(vocab[a].cn, power)
+				/ (real) train_words_pow;
+}
+
+// Reads a single word from a file, assuming space + tab + EOL to be word boundaries
+void ReadWord(char *word, FILE *fin) {
+	int a = 0, ch;
+	while (!feof(fin)) {
+		ch = fgetc(fin);
+		if (ch == 13)
+			continue;
+		if ((ch == ' ') || (ch == '\t') || (ch == '\n')) {
+			if (a > 0) {
+				if (ch == '\n')
+					ungetc(ch, fin);
+				break;
+			}
+			if (ch == '\n') {
+				strcpy(word, (char *) "</s>");
+				return;
+			} else
+				continue;
+		}
+		word[a] = ch;
+		a++;
+		if (a >= MAX_STRING - 1)
+			a--;   // Truncate too long words
+	}
+	word[a] = 0;
+}
+
+// Returns hash value of a word
+int GetWordHash(char *word) {
+	unsigned long long a, hash = 0;
+	for (a = 0; a < strlen(word); a++)
+		hash = hash * 257 + word[a];
+	hash = hash % vocab_hash_size;
+	return hash;
+}
+
+// Returns position of a word in the vocabulary; if the word is not found, returns -1
+int SearchVocab(char *word) {
+	unsigned int hash = GetWordHash(word);
+	while (1) {
+		if (vocab_hash[hash] == -1)
+			return -1;
+		if (!strcmp(word, vocab[vocab_hash[hash]].word))
+			return vocab_hash[hash];
+		hash = (hash + 1) % vocab_hash_size;
+	}
+	return -1;
+}
+
+// Reads a word and returns its index in the vocabulary
+int ReadWordIndex(FILE *fin) {
+	char word[MAX_STRING];
+	ReadWord(word, fin);
+	if (feof(fin))
+		return -1;
+	return SearchVocab(word);
+}
+
+// Adds a word to the vocabulary
+int AddWordToVocab(char *word) {
+	unsigned int hash, length = strlen(word) + 1;
+	if (length > MAX_STRING)
+		length = MAX_STRING;
+	vocab[vocab_size].word = (char *) calloc(length, sizeof(char));
+	strcpy(vocab[vocab_size].word, word);
+	vocab[vocab_size].cn = 0;
+	vocab_size++;
+	// Reallocate memory if needed
+	if (vocab_size + 2 >= vocab_max_size) {
+		vocab_max_size += 1000;
+		vocab = (struct vocab_word *) realloc(vocab,
+				vocab_max_size * sizeof(struct vocab_word));
+	}
+	hash = GetWordHash(word);
+	while (vocab_hash[hash] != -1)
+		hash = (hash + 1) % vocab_hash_size;
+	vocab_hash[hash] = vocab_size - 1;
+	return vocab_size - 1;
+}
+
+// Used later for sorting by word counts
+int VocabCompare(const void *a, const void *b) {
+	return ((struct vocab_word *) b)->cn - ((struct vocab_word *) a)->cn;
+}
+
+// Sorts the vocabulary by frequency using word counts
+void SortVocab() {
+	int a, size;
+	unsigned int hash;
+	// Sort the vocabulary and keep </s> at the first position
+	qsort(&vocab[1], vocab_size - 1, sizeof(struct vocab_word), VocabCompare);
+	for (a = 0; a < vocab_hash_size; a++)
+		vocab_hash[a] = -1;
+	size = vocab_size;
+	train_words = 0;
+	for (a = 0; a < size; a++) {
+		// Words occuring less than min_count times will be discarded from the vocab
+		if ((vocab[a].cn < min_count) && (a != 0)) {
+			vocab_size--;
+			free(vocab[a].word);
+		} else {
+			// Hash will be re-computed, as after the sorting it is not actual
+			hash = GetWordHash(vocab[a].word);
+			while (vocab_hash[hash] != -1)
+				hash = (hash + 1) % vocab_hash_size;
+			vocab_hash[hash] = a;
+			train_words += vocab[a].cn;
+		}
+	}
+	vocab = (struct vocab_word *) realloc(vocab,
+			(vocab_size + 1) * sizeof(struct vocab_word));
+	// Allocate memory for the binary tree construction
+	for (a = 0; a < vocab_size; a++) {
+		vocab[a].code = (char *) calloc(MAX_CODE_LENGTH, sizeof(char));
+		vocab[a].point = (int *) calloc(MAX_CODE_LENGTH, sizeof(int));
+	}
+}
+
+// Reduces the vocabulary by removing infrequent tokens
+void ReduceVocab() {
+	int a, b = 0;
+	unsigned int hash;
+	for (a = 0; a < vocab_size; a++)
+		if (vocab[a].cn > min_reduce) {
+			vocab[b].cn = vocab[a].cn;
+			vocab[b].word = vocab[a].word;
+			b++;
+		} else
+			free(vocab[a].word);
+	vocab_size = b;
+	for (a = 0; a < vocab_hash_size; a++)
+		vocab_hash[a] = -1;
+	for (a = 0; a < vocab_size; a++) {
+		// Hash will be re-computed, as it is not actual
+		hash = GetWordHash(vocab[a].word);
+		while (vocab_hash[hash] != -1)
+			hash = (hash + 1) % vocab_hash_size;
+		vocab_hash[hash] = a;
+	}
+	fflush(stdout);
+	min_reduce++;
+}
+
+// Create binary Huffman tree using the word counts
+// Frequent words will have short uniqe binary codes
+void CreateBinaryTree() {
+	long long a, b, i, min1i, min2i, pos1, pos2, point[MAX_CODE_LENGTH];
+	char code[MAX_CODE_LENGTH];
+	long long *count = (long long *) calloc(vocab_size * 2 + 1,
+			sizeof(long long));
+	long long *binary = (long long *) calloc(vocab_size * 2 + 1,
+			sizeof(long long));
+	long long *parent_node = (long long *) calloc(vocab_size * 2 + 1,
+			sizeof(long long));
+	for (a = 0; a < vocab_size; a++)
+		count[a] = vocab[a].cn;
+	for (a = vocab_size; a < vocab_size * 2; a++)
+		count[a] = 1e15;
+	pos1 = vocab_size - 1;
+	pos2 = vocab_size;
+	// Following algorithm constructs the Huffman tree by adding one node at a time
+	for (a = 0; a < vocab_size - 1; a++) {
+		// First, find two smallest nodes 'min1, min2'
+		if (pos1 >= 0) {
+			if (count[pos1] < count[pos2]) {
+				min1i = pos1;
+				pos1--;
+			} else {
+				min1i = pos2;
+				pos2++;
+			}
+		} else {
+			min1i = pos2;
+			pos2++;
+		}
+		if (pos1 >= 0) {
+			if (count[pos1] < count[pos2]) {
+				min2i = pos1;
+				pos1--;
+			} else {
+				min2i = pos2;
+				pos2++;
+			}
+		} else {
+			min2i = pos2;
+			pos2++;
+		}
+		count[vocab_size + a] = count[min1i] + count[min2i];
+		parent_node[min1i] = vocab_size + a;
+		parent_node[min2i] = vocab_size + a;
+		binary[min2i] = 1;
+	}
+	// Now assign binary code to each vocabulary word
+	for (a = 0; a < vocab_size; a++) {
+		b = a;
+		i = 0;
+		while (1) {
+			code[i] = binary[b];
+			point[i] = b;
+			i++;
+			b = parent_node[b];
+			if (b == vocab_size * 2 - 2)
+				break;
+		}
+		vocab[a].codelen = i;
+		vocab[a].point[0] = vocab_size - 2;
+		for (b = 0; b < i; b++) {
+			vocab[a].code[i - b - 1] = code[b];
+			vocab[a].point[i - b] = point[b] - vocab_size;
+		}
+	}
+	free(count);
+	free(binary);
+	free(parent_node);
+}
+
+void LearnVocabFromTrainFile() {
+	char word[MAX_STRING];
+	FILE *fin;
+	long long a, i;
+	for (a = 0; a < vocab_hash_size; a++)
+		vocab_hash[a] = -1;
+	fin = fopen(train_file, "rb");
+	if (fin == NULL) {
+		printf("ERROR: training data file not found!\n");
+		exit(1);
+	}
+	vocab_size = 0;
+	AddWordToVocab((char *) "</s>");
+	while (1) {
+		ReadWord(word, fin);
+		if (feof(fin))
+			break;
+		train_words++;
+		if ((debug_mode > 1) && (train_words % 100000 == 0)) {
+			printf("%lldK%c", train_words / 1000, 13);
+			fflush(stdout);
+		}
+		i = SearchVocab(word);
+		if (i == -1) {
+			a = AddWordToVocab(word);
+			vocab[a].cn = 1;
+		} else
+			vocab[i].cn++;
+		if (vocab_size > vocab_hash_size * 0.7)
+			ReduceVocab();
+	}
+	SortVocab();
+	if (debug_mode > 0) {
+		printf("Vocab size: %lld\n", vocab_size);
+		printf("Words in train file: %lld\n", train_words);
+	}
+	file_size = ftell(fin);
+	fclose(fin);
+}
+
+void SaveVocab() {
+	long long i;
+	FILE *fo = fopen(save_vocab_file, "wb");
+	for (i = 0; i < vocab_size; i++)
+		fprintf(fo, "%s %lld\n", vocab[i].word, vocab[i].cn);
+	fclose(fo);
+}
+
+void ReadVocab() {
+	long long a, i = 0;
+	char c;
+	char word[MAX_STRING];
+	FILE *fin = fopen(read_vocab_file, "rb");
+	if (fin == NULL) {
+		printf("Vocabulary file not found\n");
+		exit(1);
+	}
+	for (a = 0; a < vocab_hash_size; a++)
+		vocab_hash[a] = -1;
+	vocab_size = 0;
+	while (1) {
+		ReadWord(word, fin);
+		if (feof(fin))
+			break;
+		a = AddWordToVocab(word);
+		fscanf(fin, "%lld%c", &vocab[a].cn, &c);
+		i++;
+	}
+	SortVocab();
+	if (debug_mode > 0) {
+		printf("Vocab size: %lld\n", vocab_size);
+		printf("Words in train file: %lld\n", train_words);
+	}
+	fin = fopen(train_file, "rb");
+	if (fin == NULL) {
+		printf("ERROR: training data file not found!\n");
+		exit(1);
+	}
+	fseek(fin, 0, SEEK_END);
+	file_size = ftell(fin);
+	fclose(fin);
+}
+
+void InitClassUnigramTable() {
+	long long a, c;
+	printf("loading class unigrams \n");
+	FILE *fin = fopen(negative_classes_file, "rb");
+	if (fin == NULL) {
+		printf("ERROR: class file not found!\n");
+		exit(1);
+	}
+	word_to_group = (int *) malloc(vocab_size * sizeof(int));
+	for (a = 0; a < vocab_size; a++)
+		word_to_group[a] = -1;
+	char class[MAX_STRING];
+	char prev_class[MAX_STRING];
+	prev_class[0] = 0;
+	char word[MAX_STRING];
+	class_number = -1;
+	while (1) {
+		if (feof(fin))
+			break;
+		ReadWord(class, fin);
+		ReadWord(word, fin);
+		int word_index = SearchVocab(word);
+		if (word_index != -1) {
+			if (strcmp(class, prev_class) != 0) {
+				class_number++;
+				strcpy(prev_class, class);
+			}
+			word_to_group[word_index] = class_number;
+		}
+		ReadWord(word, fin);
+	}
+	class_number++;
+	fclose(fin);
+
+	group_to_table = (int *) malloc(table_size * class_number * sizeof(int));
+	long long train_words_pow = 0;
+	real d1, power = 0.75;
+
+	for (c = 0; c < class_number; c++) {
+		long long offset = c * table_size;
+		train_words_pow = 0;
+		for (a = 0; a < vocab_size; a++)
+			if (word_to_group[a] == c)
+				train_words_pow += pow(vocab[a].cn, power);
+		int i = 0;
+		while (word_to_group[i] != c && i < vocab_size)
+			i++;
+		d1 = pow(vocab[i].cn, power) / (real) train_words_pow;
+		for (a = 0; a < table_size; a++) {
+			//printf("index %lld , word %d\n", a, i);
+			group_to_table[offset + a] = i;
+			if (a / (real) table_size > d1) {
+				i++;
+				while (word_to_group[i] != c && i < vocab_size)
+					i++;
+				d1 += pow(vocab[i].cn, power) / (real) train_words_pow;
+			}
+			if (i >= vocab_size)
+				while (word_to_group[i] != c && i >= 0)
+					i--;
+		}
+	}
+}
+
+void SaveNet() {
+	long long a, b;
+	FILE *fnet = fopen(save_net_file, "wb");
+	if (fnet == NULL) {
+		printf("Net parameter file not found\n");
+		exit(1);
+	}
+	for (a = 0; a < vocab_size; a++)
+		for (b = 0; b < layer1_size; b++) {
+			fwrite(&syn0[a * layer1_size + b], sizeof(real), 1, fnet);
+		}
+	for (a = 0; a < window_hidden_size * window_layer_size; a++) {
+		fwrite(&syn_window_hidden[a],sizeof(real),1,fnet);
+	}
+	fclose(fnet);
+}
+
+void InitNet() {
+	long long a, b;
+	unsigned long long next_random = 1;
+	window_layer_size = layer1_size * window * 2;
+	a = posix_memalign((void **) &syn0, 128,
+			(long long) vocab_size * layer1_size * sizeof(real));
+	if (syn0 == NULL) {
+		printf("Memory allocation failed\n");
+		exit(1);
+	}
+
+	if (hs) {
+		a = posix_memalign((void **) &syn1, 128,
+				(long long) vocab_size * layer1_size * sizeof(real));
+		if (syn1 == NULL) {
+			printf("Memory allocation failed\n");
+			exit(1);
+		}
+		a = posix_memalign((void **) &syn1_window, 128,
+				(long long) vocab_size * window_layer_size * sizeof(real));
+		if (syn1_window == NULL) {
+			printf("Memory allocation failed\n");
+			exit(1);
+		}
+		a = posix_memalign((void **) &syn_hidden_word, 128,
+				(long long) vocab_size * window_hidden_size * sizeof(real));
+		if (syn_hidden_word == NULL) {
+			printf("Memory allocation failed\n");
+			exit(1);
+		}
+
+		for (a = 0; a < vocab_size; a++)
+			for (b = 0; b < layer1_size; b++)
+				syn1[a * layer1_size + b] = 0;
+		for (a = 0; a < vocab_size; a++)
+			for (b = 0; b < window_layer_size; b++)
+				syn1_window[a * window_layer_size + b] = 0;
+		for (a = 0; a < vocab_size; a++)
+			for (b = 0; b < window_hidden_size; b++)
+				syn_hidden_word[a * window_hidden_size + b] = 0;
+	}
+	if (negative > 0) {
+		a = posix_memalign((void **) &syn1neg, 128,
+				(long long) vocab_size * layer1_size * sizeof(real));
+		if (syn1neg == NULL) {
+			printf("Memory allocation failed\n");
+			exit(1);
+		}
+		a = posix_memalign((void **) &syn1neg_window, 128,
+				(long long) vocab_size * window_layer_size * sizeof(real));
+		if (syn1neg_window == NULL) {
+			printf("Memory allocation failed\n");
+			exit(1);
+		}
+		a = posix_memalign((void **) &syn_hidden_word_neg, 128,
+				(long long) vocab_size * window_hidden_size * sizeof(real));
+		if (syn_hidden_word_neg == NULL) {
+			printf("Memory allocation failed\n");
+			exit(1);
+		}
+
+		for (a = 0; a < vocab_size; a++)
+			for (b = 0; b < layer1_size; b++)
+				syn1neg[a * layer1_size + b] = 0;
+		for (a = 0; a < vocab_size; a++)
+			for (b = 0; b < window_layer_size; b++)
+				syn1neg_window[a * window_layer_size + b] = 0;
+		for (a = 0; a < vocab_size; a++)
+			for (b = 0; b < window_hidden_size; b++)
+				syn_hidden_word_neg[a * window_hidden_size + b] = 0;
+	}
+	if (nce > 0) {
+		a = posix_memalign((void **) &syn1nce, 128,
+				(long long) vocab_size * layer1_size * sizeof(real));
+		if (syn1nce == NULL) {
+			printf("Memory allocation failed\n");
+			exit(1);
+		}
+		a = posix_memalign((void **) &syn1nce_window, 128,
+				(long long) vocab_size * window_layer_size * sizeof(real));
+		if (syn1nce_window == NULL) {
+			printf("Memory allocation failed\n");
+			exit(1);
+		}
+		a = posix_memalign((void **) &syn_hidden_word_nce, 128,
+				(long long) vocab_size * window_hidden_size * sizeof(real));
+		if (syn_hidden_word_nce == NULL) {
+			printf("Memory allocation failed\n");
+			exit(1);
+		}
+
+		for (a = 0; a < vocab_size; a++)
+			for (b = 0; b < layer1_size; b++)
+				syn1nce[a * layer1_size + b] = 0;
+		for (a = 0; a < vocab_size; a++)
+			for (b = 0; b < window_layer_size; b++)
+				syn1nce_window[a * window_layer_size + b] = 0;
+		for (a = 0; a < vocab_size; a++)
+			for (b = 0; b < window_hidden_size; b++)
+				syn_hidden_word_nce[a * window_hidden_size + b] = 0;
+	}
+	if (read_net_file[0] == 0) {
+		for (a = 0; a < vocab_size; a++)
+			for (b = 0; b < layer1_size; b++) {
+				next_random = next_random * (unsigned long long) 25214903917
+						+ 11;
+				syn0[a * layer1_size + b] = (((next_random & 0xFFFF)
+						/ (real) 65536) - 0.5) / layer1_size;
+			}
+
+		a = posix_memalign((void **) &syn_window_hidden, 128,
+				window_hidden_size * window_layer_size * sizeof(real));
+		if (syn_window_hidden == NULL) {
+			printf("Memory allocation failed\n");
+			exit(1);
+		}
+		for (a = 0; a < window_hidden_size * window_layer_size; a++) {
+			next_random = next_random * (unsigned long long) 25214903917 + 11;
+			syn_window_hidden[a] = (((next_random & 0xFFFF) / (real) 65536)
+					- 0.5) / (window_hidden_size * window_layer_size);
+		}
+	}
+	else {
+		FILE *fnet = fopen(read_net_file, "rb");
+		if (fnet == NULL) {
+			printf("Net parameter file not found\n");
+			exit(1);
+		}
+		for (a = 0; a < vocab_size; a++)
+			for (b = 0; b < layer1_size; b++) {
+				fread(&syn0[a * layer1_size + b], sizeof(real), 1, fnet);
+			}
+
+		a = posix_memalign((void **) &syn_window_hidden, 128,
+				window_hidden_size * window_layer_size * sizeof(real));
+		if (syn_window_hidden == NULL) {
+			printf("Memory allocation failed\n");
+			exit(1);
+		}
+		for (a = 0; a < window_hidden_size * window_layer_size; a++) {
+			fread(&syn_window_hidden[a],sizeof(real),1,fnet);
+		}
+		fclose(fnet);
+	}
+
+	CreateBinaryTree();
+}
+
+void *TrainModelThread(void *id) {
+	long long a, b, d, cw, word, last_word, sentence_length = 0,
+			sentence_position = 0;
+	long long word_count = 0, last_word_count = 0, sen[MAX_SENTENCE_LENGTH + 1];
+	long long l1, l2, c, target, label, local_iter = iter;
+	unsigned long long next_random = (long long) id;
+	real f, g;
+	clock_t now;
+	int input_len_1 = layer1_size;
+	int window_offset = -1;
+	if (type == 2 || type == 4) {
+		input_len_1 = window_layer_size;
+	}
+	real *neu1 = (real *) calloc(input_len_1, sizeof(real));
+	real *neu1e = (real *) calloc(input_len_1, sizeof(real));
+
+	int input_len_2 = 0;
+	if (type == 4) {
+		input_len_2 = window_hidden_size;
+	}
+	real *neu2 = (real *) calloc(input_len_2, sizeof(real));
+	real *neu2e = (real *) calloc(input_len_2, sizeof(real));
+
+	FILE *fi = fopen(train_file, "rb");
+	fseek(fi, file_size / (long long) num_threads * (long long) id, SEEK_SET);
+	while (1) {
+		if (word_count - last_word_count > 10000) {
+			word_count_actual += word_count - last_word_count;
+			last_word_count = word_count;
+			if ((debug_mode > 1)) {
+				now = clock();
+				printf(
+						"%cAlpha: %f  Progress: %.2f%%  Words/thread/sec: %.2fk  ",
+						13, alpha,
+						word_count_actual / (real) (iter * train_words + 1)
+								* 100,
+						word_count_actual
+								/ ((real) (now - start + 1)
+										/ (real) CLOCKS_PER_SEC * 1000));
+				fflush(stdout);
+			}
+			alpha = starting_alpha
+					* (1 - word_count_actual / (real) (iter * train_words + 1));
+			if (alpha < starting_alpha * 0.0001)
+				alpha = starting_alpha * 0.0001;
+		}
+		if (sentence_length == 0) {
+			while (1) {
+				word = ReadWordIndex(fi);
+				if (feof(fi))
+					break;
+				if (word == -1)
+					continue;
+				word_count++;
+				if (word == 0)
+					break;
+				// The subsampling randomly discards frequent words while keeping the ranking same
+				if (sample > 0) {
+					real ran = (sqrt(vocab[word].cn / (sample * train_words))
+							+ 1) * (sample * train_words) / vocab[word].cn;
+					next_random = next_random * (unsigned long long) 25214903917
+							+ 11;
+					if (ran < (next_random & 0xFFFF) / (real) 65536)
+						continue;
+				}
+				sen[sentence_length] = word;
+				sentence_length++;
+				if (sentence_length >= MAX_SENTENCE_LENGTH)
+					break;
+			}
+			sentence_position = 0;
+		}
+		if (feof(fi) || (word_count > train_words / num_threads)) {
+			word_count_actual += word_count - last_word_count;
+			local_iter--;
+			if (local_iter == 0)
+				break;
+			word_count = 0;
+			last_word_count = 0;
+			sentence_length = 0;
+			fseek(fi, file_size / (long long) num_threads * (long long) id,
+					SEEK_SET);
+			continue;
+		}
+		word = sen[sentence_position];
+		if (word == -1)
+			continue;
+		for (c = 0; c < input_len_1; c++)
+			neu1[c] = 0;
+		for (c = 0; c < input_len_1; c++)
+			neu1e[c] = 0;
+		for (c = 0; c < input_len_2; c++)
+			neu2[c] = 0;
+		for (c = 0; c < input_len_2; c++)
+			neu2e[c] = 0;
+		next_random = next_random * (unsigned long long) 25214903917 + 11;
+		b = next_random % window;
+		if (type == 0) {  //train the cbow architecture
+			// in -> hidden
+			cw = 0;
+			for (a = b; a < window * 2 + 1 - b; a++)
+				if (a != window) {
+					c = sentence_position - window + a;
+					if (c < 0)
+						continue;
+					if (c >= sentence_length)
+						continue;
+					last_word = sen[c];
+					if (last_word == -1)
+						continue;
+					for (c = 0; c < layer1_size; c++)
+						neu1[c] += syn0[c + last_word * layer1_size];
+					cw++;
+				}
+			if (cw) {
+				for (c = 0; c < layer1_size; c++)
+					neu1[c] /= cw;
+				if (hs)
+					for (d = 0; d < vocab[word].codelen; d++) {
+						f = 0;
+						l2 = vocab[word].point[d] * layer1_size;
+						// Propagate hidden -> output
+						for (c = 0; c < layer1_size; c++)
+							f += neu1[c] * syn1[c + l2];
+						if (f <= -MAX_EXP)
+							continue;
+						else if (f >= MAX_EXP)
+							continue;
+						else
+							f = expTable[(int) ((f + MAX_EXP)
+									* (EXP_TABLE_SIZE / MAX_EXP / 2))];
+						// 'g' is the gradient multiplied by the learning rate
+						g = (1 - vocab[word].code[d] - f) * alpha;
+						// Propagate errors output -> hidden
+						for (c = 0; c < layer1_size; c++)
+							neu1e[c] += g * syn1[c + l2];
+						// Learn weights hidden -> output
+						for (c = 0; c < layer1_size; c++)
+							syn1[c + l2] += g * neu1[c];
+						if (cap == 1)
+							for (c = 0; c < layer1_size; c++)
+								capParam(syn1, c + l2);
+					}
+				// NEGATIVE SAMPLING
+				if (negative > 0)
+					for (d = 0; d < negative + 1; d++) {
+						if (d == 0) {
+							target = word;
+							label = 1;
+						} else {
+							next_random = next_random
+									* (unsigned long long) 25214903917 + 11;
+							if (word_to_group != NULL
+									&& word_to_group[word] != -1) {
+								target = word;
+								while (target == word) {
+									target = group_to_table[word_to_group[word]
+											* table_size
+											+ (next_random >> 16) % table_size];
+									next_random = next_random
+											* (unsigned long long) 25214903917
+											+ 11;
+								}
+								//printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+							} else {
+								target =
+										table[(next_random >> 16) % table_size];
+							}
+							if (target == 0)
+								target = next_random % (vocab_size - 1) + 1;
+							if (target == word)
+								continue;
+							label = 0;
+						}
+						l2 = target * layer1_size;
+						f = 0;
+						for (c = 0; c < layer1_size; c++)
+							f += neu1[c] * syn1neg[c + l2];
+						if (f > MAX_EXP)
+							g = (label - 1) * alpha;
+						else if (f < -MAX_EXP)
+							g = (label - 0) * alpha;
+						else
+							g = (label
+									- expTable[(int) ((f + MAX_EXP)
+											* (EXP_TABLE_SIZE / MAX_EXP / 2))])
+									* alpha;
+						for (c = 0; c < layer1_size; c++)
+							neu1e[c] += g * syn1neg[c + l2];
+						for (c = 0; c < layer1_size; c++)
+							syn1neg[c + l2] += g * neu1[c];
+						if (cap == 1)
+							for (c = 0; c < layer1_size; c++)
+								capParam(syn1neg, c + l2);
+					}
+				// Noise Contrastive Estimation
+				if (nce > 0)
+					for (d = 0; d < nce + 1; d++) {
+						if (d == 0) {
+							target = word;
+							label = 1;
+						} else {
+							next_random = next_random
+									* (unsigned long long) 25214903917 + 11;
+							if (word_to_group != NULL
+									&& word_to_group[word] != -1) {
+								target = word;
+								while (target == word) {
+									target = group_to_table[word_to_group[word]
+											* table_size
+											+ (next_random >> 16) % table_size];
+									next_random = next_random
+											* (unsigned long long) 25214903917
+											+ 11;
+								}
+							} else {
+								target =
+										table[(next_random >> 16) % table_size];
+							}
+							if (target == 0)
+								target = next_random % (vocab_size - 1) + 1;
+							if (target == word)
+								continue;
+							label = 0;
+						}
+						l2 = target * layer1_size;
+						f = 0;
+
+						for (c = 0; c < layer1_size; c++)
+							f += neu1[c] * syn1nce[c + l2];
+						if (f > MAX_EXP)
+							g = (label - 1) * alpha;
+						else if (f < -MAX_EXP)
+							g = (label - 0) * alpha;
+						else {
+							f = exp(f);
+							g =
+									(label
+											- f
+													/ (noise_distribution[target]
+															* nce + f)) * alpha;
+						}
+						for (c = 0; c < layer1_size; c++)
+							neu1e[c] += g * syn1nce[c + l2];
+						for (c = 0; c < layer1_size; c++)
+							syn1nce[c + l2] += g * neu1[c];
+						if (cap == 1)
+							for (c = 0; c < layer1_size; c++)
+								capParam(syn1nce, c + l2);
+					}
+				// hidden -> in
+				for (a = b; a < window * 2 + 1 - b; a++)
+					if (a != window) {
+						c = sentence_position - window + a;
+						if (c < 0)
+							continue;
+						if (c >= sentence_length)
+							continue;
+						last_word = sen[c];
+						if (last_word == -1)
+							continue;
+						for (c = 0; c < layer1_size; c++)
+							syn0[c + last_word * layer1_size] += neu1e[c];
+					}
+			}
+		} else if (type == 1) {  //train skip-gram
+			for (a = b; a < window * 2 + 1 - b; a++)
+				if (a != window) {
+					c = sentence_position - window + a;
+					if (c < 0)
+						continue;
+					if (c >= sentence_length)
+						continue;
+					last_word = sen[c];
+					if (last_word == -1)
+						continue;
+					l1 = last_word * layer1_size;
+					for (c = 0; c < layer1_size; c++)
+						neu1e[c] = 0;
+					// HIERARCHICAL SOFTMAX
+					if (hs)
+						for (d = 0; d < vocab[word].codelen; d++) {
+							f = 0;
+							l2 = vocab[word].point[d] * layer1_size;
+							// Propagate hidden -> output
+							for (c = 0; c < layer1_size; c++)
+								f += syn0[c + l1] * syn1[c + l2];
+							if (f <= -MAX_EXP)
+								continue;
+							else if (f >= MAX_EXP)
+								continue;
+							else
+								f = expTable[(int) ((f + MAX_EXP)
+										* (EXP_TABLE_SIZE / MAX_EXP / 2))];
+							// 'g' is the gradient multiplied by the learning rate
+							g = (1 - vocab[word].code[d] - f) * alpha;
+							// Propagate errors output -> hidden
+							for (c = 0; c < layer1_size; c++)
+								neu1e[c] += g * syn1[c + l2];
+							// Learn weights hidden -> output
+							for (c = 0; c < layer1_size; c++)
+								syn1[c + l2] += g * syn0[c + l1];
+							if (cap == 1)
+								for (c = 0; c < layer1_size; c++)
+									capParam(syn1, c + l2);
+						}
+					// NEGATIVE SAMPLING
+					if (negative > 0)
+						for (d = 0; d < negative + 1; d++) {
+							if (d == 0) {
+								target = word;
+								label = 1;
+							} else {
+								next_random = next_random
+										* (unsigned long long) 25214903917 + 11;
+								if (word_to_group != NULL
+										&& word_to_group[word] != -1) {
+									target = word;
+									while (target == word) {
+										target =
+												group_to_table[word_to_group[word]
+														* table_size
+														+ (next_random >> 16)
+																% table_size];
+										next_random =
+												next_random
+														* (unsigned long long) 25214903917
+														+ 11;
+									}
+									//printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+								} else {
+									target = table[(next_random >> 16)
+											% table_size];
+								}
+								if (target == 0)
+									target = next_random % (vocab_size - 1) + 1;
+								if (target == word)
+									continue;
+								label = 0;
+							}
+							l2 = target * layer1_size;
+							f = 0;
+							for (c = 0; c < layer1_size; c++)
+								f += syn0[c + l1] * syn1neg[c + l2];
+							if (f > MAX_EXP)
+								g = (label - 1) * alpha;
+							else if (f < -MAX_EXP)
+								g = (label - 0) * alpha;
+							else
+								g =
+										(label
+												- expTable[(int) ((f + MAX_EXP)
+														* (EXP_TABLE_SIZE
+																/ MAX_EXP / 2))])
+												* alpha;
+							for (c = 0; c < layer1_size; c++)
+								neu1e[c] += g * syn1neg[c + l2];
+							for (c = 0; c < layer1_size; c++)
+								syn1neg[c + l2] += g * syn0[c + l1];
+							if (cap == 1)
+								for (c = 0; c < layer1_size; c++)
+									capParam(syn1neg, c + l2);
+						}
+					//Noise Contrastive Estimation
+					if (nce > 0)
+						for (d = 0; d < nce + 1; d++) {
+							if (d == 0) {
+								target = word;
+								label = 1;
+							} else {
+								next_random = next_random
+										* (unsigned long long) 25214903917 + 11;
+								if (word_to_group != NULL
+										&& word_to_group[word] != -1) {
+									target = word;
+									while (target == word) {
+										target =
+												group_to_table[word_to_group[word]
+														* table_size
+														+ (next_random >> 16)
+																% table_size];
+										next_random =
+												next_random
+														* (unsigned long long) 25214903917
+														+ 11;
+									}
+									//printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+								} else {
+									target = table[(next_random >> 16)
+											% table_size];
+								}
+								if (target == 0)
+									target = next_random % (vocab_size - 1) + 1;
+								if (target == word)
+									continue;
+								label = 0;
+							}
+							l2 = target * layer1_size;
+							f = 0;
+							for (c = 0; c < layer1_size; c++)
+								f += syn0[c + l1] * syn1nce[c + l2];
+							if (f > MAX_EXP)
+								g = (label - 1) * alpha;
+							else if (f < -MAX_EXP)
+								g = (label - 0) * alpha;
+							else {
+								f = exp(f);
+								g = (label
+										- f
+												/ (noise_distribution[target]
+														* nce + f)) * alpha;
+							}
+							for (c = 0; c < layer1_size; c++)
+								neu1e[c] += g * syn1nce[c + l2];
+							for (c = 0; c < layer1_size; c++)
+								syn1nce[c + l2] += g * syn0[c + l1];
+							if (cap == 1)
+								for (c = 0; c < layer1_size; c++)
+									capParam(syn1nce, c + l2);
+						}
+					// Learn weights input -> hidden
+					for (c = 0; c < layer1_size; c++)
+						syn0[c + l1] += neu1e[c];
+				}
+		} else if (type == 2) { //train the cwindow architecture
+			// in -> hidden
+			cw = 0;
+			for (a = 0; a < window * 2 + 1; a++)
+				if (a != window) {
+					c = sentence_position - window + a;
+					if (c < 0)
+						continue;
+					if (c >= sentence_length)
+						continue;
+					last_word = sen[c];
+					if (last_word == -1)
+						continue;
+					window_offset = a * layer1_size;
+					if (a > window)
+						window_offset -= layer1_size;
+					for (c = 0; c < layer1_size; c++)
+						neu1[c + window_offset] += syn0[c
+								+ last_word * layer1_size];
+					cw++;
+				}
+			if (cw) {
+				if (hs)
+					for (d = 0; d < vocab[word].codelen; d++) {
+						f = 0;
+						l2 = vocab[word].point[d] * window_layer_size;
+						// Propagate hidden -> output
+						for (c = 0; c < window_layer_size; c++)
+							f += neu1[c] * syn1_window[c + l2];
+						if (f <= -MAX_EXP)
+							continue;
+						else if (f >= MAX_EXP)
+							continue;
+						else
+							f = expTable[(int) ((f + MAX_EXP)
+									* (EXP_TABLE_SIZE / MAX_EXP / 2))];
+						// 'g' is the gradient multiplied by the learning rate
+						g = (1 - vocab[word].code[d] - f) * alpha;
+						// Propagate errors output -> hidden
+						for (c = 0; c < window_layer_size; c++)
+							neu1e[c] += g * syn1_window[c + l2];
+						// Learn weights hidden -> output
+						for (c = 0; c < window_layer_size; c++)
+							syn1_window[c + l2] += g * neu1[c];
+						if (cap == 1)
+							for (c = 0; c < window_layer_size; c++)
+								capParam(syn1_window, c + l2);
+					}
+				// NEGATIVE SAMPLING
+				if (negative > 0)
+					for (d = 0; d < negative + 1; d++) {
+						if (d == 0) {
+							target = word;
+							label = 1;
+						} else {
+							next_random = next_random
+									* (unsigned long long) 25214903917 + 11;
+							if (word_to_group != NULL
+									&& word_to_group[word] != -1) {
+								target = word;
+								while (target == word) {
+									target = group_to_table[word_to_group[word]
+											* table_size
+											+ (next_random >> 16) % table_size];
+									next_random = next_random
+											* (unsigned long long) 25214903917
+											+ 11;
+								}
+								//printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+							} else {
+								target =
+										table[(next_random >> 16) % table_size];
+							}
+							if (target == 0)
+								target = next_random % (vocab_size - 1) + 1;
+							if (target == word)
+								continue;
+							label = 0;
+						}
+						l2 = target * window_layer_size;
+						f = 0;
+						for (c = 0; c < window_layer_size; c++)
+							f += neu1[c] * syn1neg_window[c + l2];
+						if (f > MAX_EXP)
+							g = (label - 1) * alpha;
+						else if (f < -MAX_EXP)
+							g = (label - 0) * alpha;
+						else
+							g = (label
+									- expTable[(int) ((f + MAX_EXP)
+											* (EXP_TABLE_SIZE / MAX_EXP / 2))])
+									* alpha;
+						for (c = 0; c < window_layer_size; c++)
+							neu1e[c] += g * syn1neg_window[c + l2];
+						for (c = 0; c < window_layer_size; c++)
+							syn1neg_window[c + l2] += g * neu1[c];
+						if (cap == 1)
+							for (c = 0; c < window_layer_size; c++)
+								capParam(syn1neg_window, c + l2);
+					}
+				// Noise Contrastive Estimation
+				if (nce > 0)
+					for (d = 0; d < nce + 1; d++) {
+						if (d == 0) {
+							target = word;
+							label = 1;
+						} else {
+							next_random = next_random
+									* (unsigned long long) 25214903917 + 11;
+							if (word_to_group != NULL
+									&& word_to_group[word] != -1) {
+								target = word;
+								while (target == word) {
+									target = group_to_table[word_to_group[word]
+											* table_size
+											+ (next_random >> 16) % table_size];
+									next_random = next_random
+											* (unsigned long long) 25214903917
+											+ 11;
+								}
+								//printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+							} else {
+								target =
+										table[(next_random >> 16) % table_size];
+							}
+							if (target == 0)
+								target = next_random % (vocab_size - 1) + 1;
+							if (target == word)
+								continue;
+							label = 0;
+						}
+						l2 = target * window_layer_size;
+						f = 0;
+						for (c = 0; c < window_layer_size; c++)
+							f += neu1[c] * syn1nce_window[c + l2];
+						if (f > MAX_EXP)
+							g = (label - 1) * alpha;
+						else if (f < -MAX_EXP)
+							g = (label - 0) * alpha;
+						else {
+							f = exp(f);
+							g =
+									(label
+											- f
+													/ (noise_distribution[target]
+															* nce + f)) * alpha;
+						}
+						for (c = 0; c < window_layer_size; c++)
+							neu1e[c] += g * syn1nce_window[c + l2];
+						for (c = 0; c < window_layer_size; c++)
+							syn1nce_window[c + l2] += g * neu1[c];
+						if (cap == 1)
+							for (c = 0; c < window_layer_size; c++)
+								capParam(syn1nce_window, c + l2);
+					}
+				// hidden -> in
+				for (a = 0; a < window * 2 + 1; a++)
+					if (a != window) {
+						c = sentence_position - window + a;
+						if (c < 0)
+							continue;
+						if (c >= sentence_length)
+							continue;
+						last_word = sen[c];
+						if (last_word == -1)
+							continue;
+						window_offset = a * layer1_size;
+						if (a > window)
+							window_offset -= layer1_size;
+						for (c = 0; c < layer1_size; c++)
+							syn0[c + last_word * layer1_size] += neu1e[c
+									+ window_offset];
+					}
+			}
+		} else if (type == 3) {  //train structured skip-gram
+			for (a = 0; a < window * 2 + 1; a++)
+				if (a != window) {
+					c = sentence_position - window + a;
+					if (c < 0)
+						continue;
+					if (c >= sentence_length)
+						continue;
+					last_word = sen[c];
+					if (last_word == -1)
+						continue;
+					l1 = last_word * layer1_size;
+					window_offset = a * layer1_size;
+					if (a > window)
+						window_offset -= layer1_size;
+					for (c = 0; c < layer1_size; c++)
+						neu1e[c] = 0;
+					// HIERARCHICAL SOFTMAX
+					if (hs)
+						for (d = 0; d < vocab[word].codelen; d++) {
+							f = 0;
+							l2 = vocab[word].point[d] * window_layer_size;
+							// Propagate hidden -> output
+							for (c = 0; c < layer1_size; c++)
+								f += syn0[c + l1]
+										* syn1_window[c + l2 + window_offset];
+							if (f <= -MAX_EXP)
+								continue;
+							else if (f >= MAX_EXP)
+								continue;
+							else
+								f = expTable[(int) ((f + MAX_EXP)
+										* (EXP_TABLE_SIZE / MAX_EXP / 2))];
+							// 'g' is the gradient multiplied by the learning rate
+							g = (1 - vocab[word].code[d] - f) * alpha;
+							// Propagate errors output -> hidden
+							for (c = 0; c < layer1_size; c++)
+								neu1e[c] += g
+										* syn1_window[c + l2 + window_offset];
+							// Learn weights hidden -> output
+							for (c = 0; c < layer1_size; c++)
+								syn1[c + l2 + window_offset] += g
+										* syn0[c + l1];
+							if (cap == 1)
+								for (c = 0; c < layer1_size; c++)
+									capParam(syn1, c + l2 + window_offset);
+						}
+					// NEGATIVE SAMPLING
+					if (negative > 0)
+						for (d = 0; d < negative + 1; d++) {
+							if (d == 0) {
+								target = word;
+								label = 1;
+							} else {
+								next_random = next_random
+										* (unsigned long long) 25214903917 + 11;
+								if (word_to_group != NULL
+										&& word_to_group[word] != -1) {
+									target = word;
+									while (target == word) {
+										target =
+												group_to_table[word_to_group[word]
+														* table_size
+														+ (next_random >> 16)
+																% table_size];
+										next_random =
+												next_random
+														* (unsigned long long) 25214903917
+														+ 11;
+									}
+									//printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+								} else {
+									target = table[(next_random >> 16)
+											% table_size];
+								}
+								if (target == 0)
+									target = next_random % (vocab_size - 1) + 1;
+								if (target == word)
+									continue;
+								label = 0;
+							}
+							l2 = target * window_layer_size;
+							f = 0;
+							for (c = 0; c < layer1_size; c++)
+								f +=
+										syn0[c + l1]
+												* syn1neg_window[c + l2
+														+ window_offset];
+							if (f > MAX_EXP)
+								g = (label - 1) * alpha;
+							else if (f < -MAX_EXP)
+								g = (label - 0) * alpha;
+							else
+								g =
+										(label
+												- expTable[(int) ((f + MAX_EXP)
+														* (EXP_TABLE_SIZE
+																/ MAX_EXP / 2))])
+												* alpha;
+							for (c = 0; c < layer1_size; c++)
+								neu1e[c] +=
+										g
+												* syn1neg_window[c + l2
+														+ window_offset];
+							for (c = 0; c < layer1_size; c++)
+								syn1neg_window[c + l2 + window_offset] += g
+										* syn0[c + l1];
+							if (cap == 1)
+								for (c = 0; c < layer1_size; c++)
+									capParam(syn1neg_window,
+											c + l2 + window_offset);
+						}
+					// Noise Constrastive Estimation
+					if (nce > 0)
+						for (d = 0; d < nce + 1; d++) {
+							if (d == 0) {
+								target = word;
+								label = 1;
+							} else {
+								next_random = next_random
+										* (unsigned long long) 25214903917 + 11;
+								if (word_to_group != NULL
+										&& word_to_group[word] != -1) {
+									target = word;
+									while (target == word) {
+										target =
+												group_to_table[word_to_group[word]
+														* table_size
+														+ (next_random >> 16)
+																% table_size];
+										next_random =
+												next_random
+														* (unsigned long long) 25214903917
+														+ 11;
+									}
+									//printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+								} else {
+									target = table[(next_random >> 16)
+											% table_size];
+								}
+								if (target == 0)
+									target = next_random % (vocab_size - 1) + 1;
+								if (target == word)
+									continue;
+								label = 0;
+							}
+							l2 = target * window_layer_size;
+							f = 0;
+							for (c = 0; c < layer1_size; c++)
+								f +=
+										syn0[c + l1]
+												* syn1nce_window[c + l2
+														+ window_offset];
+							if (f > MAX_EXP)
+								g = (label - 1) * alpha;
+							else if (f < -MAX_EXP)
+								g = (label - 0) * alpha;
+							else {
+								f = exp(f);
+								g = (label
+										- f
+												/ (noise_distribution[target]
+														* nce + f)) * alpha;
+							}
+							for (c = 0; c < layer1_size; c++)
+								neu1e[c] +=
+										g
+												* syn1nce_window[c + l2
+														+ window_offset];
+							for (c = 0; c < layer1_size; c++)
+								syn1nce_window[c + l2 + window_offset] += g
+										* syn0[c + l1];
+							if (cap == 1)
+								for (c = 0; c < layer1_size; c++)
+									capParam(syn1nce_window,
+											c + l2 + window_offset);
+						}
+					// Learn weights input -> hidden
+					for (c = 0; c < layer1_size; c++) {
+						syn0[c + l1] += neu1e[c];
+						if (syn0[c + l1] > 50)
+							syn0[c + l1] = 50;
+						if (syn0[c + l1] < -50)
+							syn0[c + l1] = -50;
+					}
+				}
+		} else if (type == 4) { //training senna
+			// in -> hidden
+			cw = 0;
+			for (a = 0; a < window * 2 + 1; a++)
+				if (a != window) {
+					c = sentence_position - window + a;
+					if (c < 0)
+						continue;
+					if (c >= sentence_length)
+						continue;
+					last_word = sen[c];
+					if (last_word == -1)
+						continue;
+					window_offset = a * layer1_size;
+					if (a > window)
+						window_offset -= layer1_size;
+					for (c = 0; c < layer1_size; c++)
+						neu1[c + window_offset] += syn0[c
+								+ last_word * layer1_size];
+					cw++;
+				}
+			if (cw) {
+				for (a = 0; a < window_hidden_size; a++) {
+					c = a * window_layer_size;
+					for (b = 0; b < window_layer_size; b++) {
+						neu2[a] += syn_window_hidden[c + b] * neu1[b];
+					}
+				}
+				if (hs)
+					for (d = 0; d < vocab[word].codelen; d++) {
+						f = 0;
+						l2 = vocab[word].point[d] * window_hidden_size;
+						// Propagate hidden -> output
+						for (c = 0; c < window_hidden_size; c++)
+							f += hardTanh(neu2[c]) * syn_hidden_word[c + l2];
+						if (f <= -MAX_EXP)
+							continue;
+						else if (f >= MAX_EXP)
+							continue;
+						else
+							f = expTable[(int) ((f + MAX_EXP)
+									* (EXP_TABLE_SIZE / MAX_EXP / 2))];
+						// 'g' is the gradient multiplied by the learning rate
+						g = (1 - vocab[word].code[d] - f) * alpha;
+						// Propagate errors output -> hidden
+						for (c = 0; c < window_hidden_size; c++)
+							neu2e[c] += dHardTanh(neu2[c], g) * g
+									* syn_hidden_word[c + l2];
+						// Learn weights hidden -> output
+						for (c = 0; c < window_hidden_size; c++)
+							syn_hidden_word[c + l2] += dHardTanh(neu2[c], g) * g
+									* neu2[c];
+					}
+				// NEGATIVE SAMPLING
+				if (negative > 0)
+					for (d = 0; d < negative + 1; d++) {
+						if (d == 0) {
+							target = word;
+							label = 1;
+						} else {
+							next_random = next_random
+									* (unsigned long long) 25214903917 + 11;
+							if (word_to_group != NULL
+									&& word_to_group[word] != -1) {
+								target = word;
+								while (target == word) {
+									target = group_to_table[word_to_group[word]
+											* table_size
+											+ (next_random >> 16) % table_size];
+									next_random = next_random
+											* (unsigned long long) 25214903917
+											+ 11;
+								}
+								//printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+							} else {
+								target =
+										table[(next_random >> 16) % table_size];
+							}
+							if (target == 0)
+								target = next_random % (vocab_size - 1) + 1;
+							if (target == word)
+								continue;
+							label = 0;
+						}
+						l2 = target * window_hidden_size;
+						f = 0;
+						for (c = 0; c < window_hidden_size; c++)
+							f += hardTanh(neu2[c])
+									* syn_hidden_word_neg[c + l2];
+						if (f > MAX_EXP)
+							g = (label - 1) * alpha / negative;
+						else if (f < -MAX_EXP)
+							g = (label - 0) * alpha / negative;
+						else
+							g = (label
+									- expTable[(int) ((f + MAX_EXP)
+											* (EXP_TABLE_SIZE / MAX_EXP / 2))])
+									* alpha / negative;
+						for (c = 0; c < window_hidden_size; c++)
+							neu2e[c] += dHardTanh(neu2[c], g) * g
+									* syn_hidden_word_neg[c + l2];
+						for (c = 0; c < window_hidden_size; c++)
+							syn_hidden_word_neg[c + l2] += dHardTanh(neu2[c], g)
+									* g * neu2[c];
+					}
+				for (a = 0; a < window_hidden_size; a++)
+					for (b = 0; b < window_layer_size; b++)
+						neu1e[b] += neu2e[a]
+								* syn_window_hidden[a * window_layer_size + b];
+				for (a = 0; a < window_hidden_size; a++)
+					for (b = 0; b < window_layer_size; b++)
+						syn_window_hidden[a * window_layer_size + b] += neu2e[a]
+								* neu1[b];
+				// hidden -> in
+				for (a = 0; a < window * 2 + 1; a++)
+					if (a != window) {
+						c = sentence_position - window + a;
+						if (c < 0)
+							continue;
+						if (c >= sentence_length)
+							continue;
+						last_word = sen[c];
+						if (last_word == -1)
+							continue;
+						window_offset = a * layer1_size;
+						if (a > window)
+							window_offset -= layer1_size;
+						for (c = 0; c < layer1_size; c++)
+							syn0[c + last_word * layer1_size] += neu1e[c
+									+ window_offset];
+					}
+			}
+		} else {
+			printf("unknown type %i", type);
+			exit(0);
+		}
+		sentence_position++;
+		if (sentence_position >= sentence_length) {
+			sentence_length = 0;
+			continue;
+		}
+	}
+	fclose(fi);
+	free(neu1);
+	free(neu1e);
+	pthread_exit(NULL);
+}
+
+void TrainModel() {
+	long a, b, c, d;
+	FILE *fo;
+	pthread_t *pt = (pthread_t *) malloc(num_threads * sizeof(pthread_t));
+	printf("Starting training using file %s\n", train_file);
+	starting_alpha = alpha;
+	if (read_vocab_file[0] != 0)
+		ReadVocab();
+	else
+		LearnVocabFromTrainFile();
+	if (save_vocab_file[0] != 0)
+		SaveVocab();
+	if (output_file[0] == 0)
+		return;
+	InitNet();
+	if (negative > 0 || nce > 0)
+		InitUnigramTable();
+	if (negative_classes_file[0] != 0)
+		InitClassUnigramTable();
+	start = clock();
+	for (a = 0; a < num_threads; a++)
+		pthread_create(&pt[a], NULL, TrainModelThread, (void *) a);
+	for (a = 0; a < num_threads; a++)
+		pthread_join(pt[a], NULL);
+	fo = fopen(output_file, "wb");
+	if (classes == 0) {
+		// Save the word vectors
+		fprintf(fo, "%lld %lld\n", vocab_size, layer1_size);
+		for (a = 0; a < vocab_size; a++) {
+			fprintf(fo, "%s ", vocab[a].word);
+			if (binary)
+				for (b = 0; b < layer1_size; b++)
+					fwrite(&syn0[a * layer1_size + b], sizeof(real), 1, fo);
+			else
+				for (b = 0; b < layer1_size; b++)
+					fprintf(fo, "%lf ", syn0[a * layer1_size + b]);
+			fprintf(fo, "\n");
+		}
+	} else {
+		// Run K-means on the word vectors
+		int clcn = classes, iter = 10, closeid;
+		int *centcn = (int *) malloc(classes * sizeof(int));
+		int *cl = (int *) calloc(vocab_size, sizeof(int));
+		real closev, x;
+		real *cent = (real *) calloc(classes * layer1_size, sizeof(real));
+		for (a = 0; a < vocab_size; a++)
+			cl[a] = a % clcn;
+		for (a = 0; a < iter; a++) {
+			for (b = 0; b < clcn * layer1_size; b++)
+				cent[b] = 0;
+			for (b = 0; b < clcn; b++)
+				centcn[b] = 1;
+			for (c = 0; c < vocab_size; c++) {
+				for (d = 0; d < layer1_size; d++)
+					cent[layer1_size * cl[c] + d] += syn0[c * layer1_size + d];
+				centcn[cl[c]]++;
+			}
+			for (b = 0; b < clcn; b++) {
+				closev = 0;
+				for (c = 0; c < layer1_size; c++) {
+					cent[layer1_size * b + c] /= centcn[b];
+					closev += cent[layer1_size * b + c]
+							* cent[layer1_size * b + c];
+				}
+				closev = sqrt(closev);
+				for (c = 0; c < layer1_size; c++)
+					cent[layer1_size * b + c] /= closev;
+			}
+			for (c = 0; c < vocab_size; c++) {
+				closev = -10;
+				closeid = 0;
+				for (d = 0; d < clcn; d++) {
+					x = 0;
+					for (b = 0; b < layer1_size; b++)
+						x += cent[layer1_size * d + b]
+								* syn0[c * layer1_size + b];
+					if (x > closev) {
+						closev = x;
+						closeid = d;
+					}
+				}
+				cl[c] = closeid;
+			}
+		}
+		// Save the K-means classes
+		for (a = 0; a < vocab_size; a++)
+			fprintf(fo, "%s %d\n", vocab[a].word, cl[a]);
+		free(centcn);
+		free(cent);
+		free(cl);
+	}
+	fclose(fo);
+	if (save_net_file[0] != 0)
+		SaveNet();
+}
+
+int ArgPos(char *str, int argc, char **argv) {
+	int a;
+	for (a = 1; a < argc; a++)
+		if (!strcmp(str, argv[a])) {
+			if (a == argc - 1) {
+				printf("Argument missing for %s\n", str);
+				exit(1);
+			}
+			return a;
+		}
+	return -1;
+}
+
+int main(int argc, char **argv) {
+	int i;
+	if (argc == 1) {
+		printf("WORD VECTOR estimation toolkit v 0.1c\n\n");
+		printf("Options:\n");
+		printf("Parameters for training:\n");
+		printf("\t-train <file>\n");
+		printf("\t\tUse text data from <file> to train the model\n");
+		printf("\t-output <file>\n");
+		printf(
+				"\t\tUse <file> to save the resulting word vectors / word clusters\n");
+		printf("\t-size <int>\n");
+		printf("\t\tSet size of word vectors; default is 100\n");
+		printf("\t-window <int>\n");
+		printf("\t\tSet max skip length between words; default is 5\n");
+		printf("\t-sample <float>\n");
+		printf(
+				"\t\tSet threshold for occurrence of words. Those that appear with higher frequency in the training data\n");
+		printf(
+				"\t\twill be randomly down-sampled; default is 1e-3, useful range is (0, 1e-5)\n");
+		printf("\t-hs <int>\n");
+		printf("\t\tUse Hierarchical Softmax; default is 0 (not used)\n");
+		printf("\t-negative <int>\n");
+		printf(
+				"\t\tNumber of negative examples; default is 5, common values are 3 - 10 (0 = not used)\n");
+		printf("\t-negative-classes <file>\n");
+		printf("\t\tNegative classes to sample from\n");
+		printf("\t-nce <int>\n");
+		printf(
+				"\t\tNumber of negative examples for nce; default is 0, common values are 3 - 10 (0 = not used)\n");
+		printf("\t-threads <int>\n");
+		printf("\t\tUse <int> threads (default 12)\n");
+		printf("\t-iter <int>\n");
+		printf("\t\tRun more training iterations (default 5)\n");
+		printf("\t-min-count <int>\n");
+		printf(
+				"\t\tThis will discard words that appear less than <int> times; default is 5\n");
+		printf("\t-alpha <float>\n");
+		printf(
+				"\t\tSet the starting learning rate; default is 0.025 for skip-gram and 0.05 for CBOW\n");
+		printf("\t-classes <int>\n");
+		printf(
+				"\t\tOutput word classes rather than word vectors; default number of classes is 0 (vectors are written)\n");
+		printf("\t-debug <int>\n");
+		printf(
+				"\t\tSet the debug mode (default = 2 = more info during training)\n");
+		printf("\t-binary <int>\n");
+		printf(
+				"\t\tSave the resulting vectors in binary moded; default is 0 (off)\n");
+		printf("\t-save-vocab <file>\n");
+		printf("\t\tThe vocabulary will be saved to <file>\n");
+		printf("\t-read-vocab <file>\n");
+		printf(
+				"\t\tThe vocabulary will be read from <file>, not constructed from the training data\n");
+		printf("\t-read-net <file>\n");
+		printf(
+				"\t\tThe net parameters will be read from <file>, not initialized randomly\n");
+		printf("\t-save-net <file>\n");
+		printf("\t\tThe net parameters will be saved to <file>\n");
+		printf("\t-type <int>\n");
+		printf(
+				"\t\tType of embeddings (0 for cbow, 1 for skipngram, 2 for cwindow, 3 for structured skipngram, 4 for senna type)\n");
+		printf("\t-cap <int>\n");
+		printf(
+				"\t\tlimit the parameter values to the range [-50, 50]; default is 0 (off)\n");
+		printf("\nExamples:\n");
+		printf(
+				"./word2vec -train data.txt -output vec.txt -size 200 -window 5 -sample 1e-4 -negative 5 -hs 0 -binary 0 -type 1 -iter 3\n\n");
+		return 0;
+	}
+	output_file[0] = 0;
+	save_vocab_file[0] = 0;
+	read_vocab_file[0] = 0;
+	save_net_file[0] = 0;
+	read_net_file[0] = 0;
+	negative_classes_file[0] = 0;
+	if ((i = ArgPos((char *) "-size", argc, argv)) > 0)
+		layer1_size = atoi(argv[i + 1]);
+	if ((i = ArgPos((char *) "-train", argc, argv)) > 0)
+		strcpy(train_file, argv[i + 1]);
+	if ((i = ArgPos((char *) "-save-vocab", argc, argv)) > 0)
+		strcpy(save_vocab_file, argv[i + 1]);
+	if ((i = ArgPos((char *) "-read-vocab", argc, argv)) > 0)
+		strcpy(read_vocab_file, argv[i + 1]);
+	if ((i = ArgPos((char *) "-save-net", argc, argv)) > 0)
+		strcpy(save_net_file, argv[i + 1]);
+	if ((i = ArgPos((char *) "-read-net", argc, argv)) > 0)
+		strcpy(read_net_file, argv[i + 1]);
+	if ((i = ArgPos((char *) "-debug", argc, argv)) > 0)
+		debug_mode = atoi(argv[i + 1]);
+	if ((i = ArgPos((char *) "-binary", argc, argv)) > 0)
+		binary = atoi(argv[i + 1]);
+	if ((i = ArgPos((char *) "-type", argc, argv)) > 0)
+		type = atoi(argv[i + 1]);
+	if ((i = ArgPos((char *) "-output", argc, argv)) > 0)
+		strcpy(output_file, argv[i + 1]);
+	if ((i = ArgPos((char *) "-window", argc, argv)) > 0)
+		window = atoi(argv[i + 1]);
+	if ((i = ArgPos((char *) "-sample", argc, argv)) > 0)
+		sample = atof(argv[i + 1]);
+	if ((i = ArgPos((char *) "-hs", argc, argv)) > 0)
+		hs = atoi(argv[i + 1]);
+	if ((i = ArgPos((char *) "-negative", argc, argv)) > 0)
+		negative = atoi(argv[i + 1]);
+	if ((i = ArgPos((char *) "-negative-classes", argc, argv)) > 0)
+		strcpy(negative_classes_file, argv[i + 1]);
+	if ((i = ArgPos((char *) "-nce", argc, argv)) > 0)
+		nce = atoi(argv[i + 1]);
+	if ((i = ArgPos((char *) "-threads", argc, argv)) > 0)
+		num_threads = atoi(argv[i + 1]);
+	if ((i = ArgPos((char *) "-iter", argc, argv)) > 0)
+		iter = atoi(argv[i + 1]);
+	if ((i = ArgPos((char *) "-min-count", argc, argv)) > 0)
+		min_count = atoi(argv[i + 1]);
+	if ((i = ArgPos((char *) "-classes", argc, argv)) > 0)
+		classes = atoi(argv[i + 1]);
+	if ((i = ArgPos((char *) "-cap", argc, argv)) > 0)
+		cap = atoi(argv[i + 1]);
+	if (type == 0 || type == 2 || type == 4)
+		alpha = 0.05;
+	if ((i = ArgPos((char *) "-alpha", argc, argv)) > 0)
+		alpha = atof(argv[i + 1]);
+	vocab = (struct vocab_word *) calloc(vocab_max_size,
+			sizeof(struct vocab_word));
+	vocab_hash = (int *) calloc(vocab_hash_size, sizeof(int));
+	expTable = (real *) malloc((EXP_TABLE_SIZE + 1) * sizeof(real));
+	for (i = 0; i < EXP_TABLE_SIZE; i++) {
+		expTable[i] = exp((i / (real) EXP_TABLE_SIZE * 2 - 1) * MAX_EXP); // Precompute the exp() table
+		expTable[i] = expTable[i] / (expTable[i] + 1); // Precompute f(x) = x / (x + 1)
+	}
+	TrainModel();
+	return 0;
+}
+
diff --git a/wordless2vec.c b/wordless2vec.c
new file mode 100644
index 0000000..e68bbee
--- /dev/null
+++ b/wordless2vec.c
@@ -0,0 +1,1696 @@
+//  Copyright 2013 Google Inc. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <pthread.h>
+
+#define MAX_STRING 100
+#define EXP_TABLE_SIZE 1000
+#define MAX_EXP 6
+#define MAX_SENTENCE_LENGTH 1000
+#define MAX_CODE_LENGTH 40
+
+const int vocab_hash_size = 30000000;  // Maximum 30 * 0.7 = 21M words in the vocabulary
+
+typedef float real;                    // Precision of float numbers
+
+struct vocab_word {
+  long long cn;
+  int *point;
+  char *word, *code, codelen;
+};
+
+char train_file[MAX_STRING], output_file[MAX_STRING];
+char save_vocab_file[MAX_STRING], read_vocab_file[MAX_STRING];
+struct vocab_word *vocab;
+int binary = 0, type = 1, debug_mode = 2, window = 5, min_count = 5, num_threads = 12, min_reduce = 1;
+int *vocab_hash;
+long long vocab_max_size = 1000, vocab_size = 0, layer1_size = 100;
+long long train_words = 0, word_count_actual = 0, iter = 5, file_size = 0, classes = 0;
+real alpha = 0.025, starting_alpha, sample = 1e-3;
+real *syn0, *syn1, *syn1neg, *expTable, *tanhTable;
+clock_t start;
+
+real *syn1_window, *syn1neg_window;
+int window_offset, window_layer_size;
+
+int window_hidden_size = 500; 
+real *syn_window_hidden, *syn_hidden_word, *syn_hidden_word_neg; 
+
+int hs = 0, negative = 5;
+const int table_size = 1e8;
+int *table;
+
+//constrastive negative sampling
+char negative_classes_file[MAX_STRING];
+int *word_to_group;
+int *group_to_table; //group_size*table_size
+int class_number;
+
+//char table 
+int rep = 0;
+#define C_MAX_CODE 65536
+int c_state_size = 5;
+int c_cell_size = 5;
+int c_proj_size = 3;
+int c_params_number;
+int c_lstm_params_number;
+real *c_lookup;
+
+//char lstm params
+real *f_init_state;
+real *f_init_cell;
+real *b_init_state;
+real *b_init_cell;
+real *f_b_params;
+
+//short term memory
+real*syn0_initial;
+real*syn0_in_memory;
+
+int batch_size = 100;
+
+void printStates(real*states, int start){
+	int s;
+	printf("igate ");	
+	for(s = 0; s < c_state_size; s++){ printf("%f ", states[start++]);} printf("\n");
+	printf("fgate ");	
+	for(s = 0; s < c_state_size; s++){ printf("%f ", states[start++]);} printf("\n");
+	printf("c + tanh ");	
+	for(s = 0; s < c_state_size; s++){ printf("%f ", states[start++]);} printf("\n");
+	printf("cgate ");	
+	for(s = 0; s < c_state_size; s++){ printf("%f ", states[start++]);} printf("\n");
+	printf("ogate ");	
+	for(s = 0; s < c_state_size; s++){ printf("%f ", states[start++]);} printf("\n");
+	printf("cgate + tanh ");	
+	for(s = 0; s < c_state_size; s++){ printf("%f ", states[start++]);} printf("\n");
+	printf("state ");	
+	for(s = 0; s < c_state_size; s++){ printf("%f ", states[start++]);} printf("\n");
+
+}
+
+void lstmForwardBlock(real *chars, int char_start, real*states, int next_start, int p){
+	int i,s,si,sf,sc,sct,sctt,so,s1=next_start;	
+	int prev_cell_start = s1 - c_state_size*4;
+	int prev_state_start = s1 - c_state_size;
+	if(states[prev_cell_start]==0){
+//		printf("crap! cell is zero\n");		
+	}
+	if(states[prev_state_start]==0){
+//		printf("crap! state is zero\n");		
+	}
+	if(states[s1]!=0){
+//		printf("crap! start not zero\n");
+	}
+	//igate
+	si = s1;
+	for(s = 0; s < c_state_size; s++){
+		for(i = 0; i < c_proj_size; i++){
+			states[s1]+=chars[char_start+i]*f_b_params[p++];
+		}
+		for(i = 0; i < c_cell_size; i++){
+			states[s1]+=states[prev_cell_start+i]*f_b_params[p++];
+		}
+		for(i = 0; i < c_state_size; i++){
+			states[s1]+=states[prev_state_start+i]*f_b_params[p++];
+		}
+		states[s1]+=f_b_params[p++];
+		if(states[s1]>MAX_EXP){
+			states[s1]=1;
+		}
+		else if(states[s1]<-MAX_EXP){
+			states[s1]=0;
+		}
+		else{
+			states[s1] = expTable[(int)((states[s1] + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+		}
+		s1++;		
+	}
+	
+	//fgate
+	sf=s1;
+	for(s = 0; s < c_state_size; s++){
+		for(i = 0; i < c_proj_size; i++){
+			states[s1]+=chars[char_start+i]*f_b_params[p++];
+		}
+		for(i = 0; i < c_cell_size; i++){
+			states[s1]+=states[prev_cell_start+i]*f_b_params[p++];
+		}
+		for(i = 0; i < c_state_size; i++){
+			states[s1]+=states[prev_state_start+i]*f_b_params[p++];
+		}
+		states[s1]+=f_b_params[p++];
+		if(states[s1]>MAX_EXP){
+			states[s1]=1;
+		}
+		else if(states[s1]<-MAX_EXP){
+			states[s1]=0;
+		}
+		else{
+			states[s1] = expTable[(int)((states[s1] + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+		}
+		s1++;
+	}
+	
+	//c + tanh
+	sct=s1;
+	for(s = 0; s < c_state_size; s++){
+		for(i = 0; i < c_proj_size; i++){
+			states[s1]+=chars[char_start+i]*f_b_params[p++];
+		}
+		for(i = 0; i < c_state_size; i++){
+			states[s1]+=states[prev_state_start+i]*f_b_params[p++];
+		}
+		states[s1]+=f_b_params[p++];
+		if(states[s1]>MAX_EXP){
+			states[s1]=1;
+		}
+		else if(states[s1]<-MAX_EXP){
+			states[s1]=-1;
+		}
+		else{
+			states[s1] = tanhTable[(int)((states[s1] + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+		}
+		s1++;
+	}
+	
+	//cgate
+	sc=s1;
+	for(s = 0; s < c_state_size; s++){
+		states[s1]+=states[sct+s]*states[si+s]+states[sf+s]*states[prev_cell_start+s];
+		s1++;
+	}
+	
+	//ogate
+	so=s1;
+	for(s = 0; s < c_state_size; s++){
+		for(i = 0; i < c_proj_size; i++){
+			states[s1]+=chars[char_start+i]*f_b_params[p++];
+		}
+		for(i = 0; i < c_cell_size; i++){
+			states[s1]+=states[sc+s]*f_b_params[p++];
+		}
+		for(i = 0; i < c_state_size; i++){
+			states[s1]+=states[prev_state_start+i]*f_b_params[p++];
+		}
+		states[s1]+=f_b_params[p++];
+		if(states[s1]>MAX_EXP){
+			states[s1]=1;
+		}
+		else if(states[s1]<-MAX_EXP){
+			states[s1]=0;
+		}
+		else{
+			states[s1] = expTable[(int)((states[s1] + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+		}
+		s1++;		
+	}
+	
+	//cgate + tan
+	sctt = s1;
+	for(s = 0; s < c_state_size; s++){
+		if(states[sc+s]>MAX_EXP){
+			states[s1]=1;
+		}
+		else if(states[sc+s]<-MAX_EXP){
+			states[s1]=-1;
+		}
+		else{
+			states[s1] = tanhTable[(int)((states[sc+s] + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+		}
+		s1++;
+	}
+	
+	//next state
+	if(states[s1]!=0){
+		printf("crap! end not zero\n");
+	}
+	for(s = 0; s < c_state_size; s++){
+		states[s1] = states[sctt+s] * states[so+s];
+		s1++;
+	}
+	
+	
+}
+
+void lstmBackwardBlock(real *chars, int char_start, real*states, int next_start, int pStart, real*chars_e, real*states_e, real*lstm_params_e){
+	int p=pStart+c_lstm_params_number-1;
+	int i,s,si,sf,sc,sct,sctt,so,s1=next_start+c_state_size*7-1;
+	int prev_cell_start = next_start - c_state_size*4;
+	int prev_state_start = next_start - c_state_size;
+
+	real e;
+	si = next_start;
+	sf = next_start + c_state_size;
+	sct = next_start + c_state_size*2;
+	sc = next_start + c_state_size*3;
+	so = next_start + c_state_size*4;
+	sctt = next_start + c_state_size*5;
+	
+	//next state 
+	for(s = c_state_size-1; s >= 0; s--){
+		states_e[sctt+s] += states_e[s1]*states[so+s];
+		states_e[so+s] += states_e[s1]*states[sctt+s];
+		s1--;
+	}	
+	
+	
+	//cgate + tan	
+	for(s = c_state_size-1; s >= 0; s--){
+		states_e[sc+s] += states_e[s1]*(1-states[s1]*states[s1]);		
+		s1--;
+	}
+	
+	//ogate
+	for(s = c_state_size-1; s >= 0; s--){
+		e = states[s1]*(1-states[s1])*states_e[s1];
+		for(i = c_proj_size-1; i >= 0; i--){
+			chars_e[char_start+i] += e*f_b_params[p];
+			lstm_params_e[p--] += e*chars[char_start+i];
+		}
+
+		for(i = c_cell_size-1; i >= 0; i--){
+			states_e[sc+s]+=e*f_b_params[p];
+			lstm_params_e[p--] += e*states[sc+s];
+		}
+		for(i = c_state_size-1; i >= 0; i--){
+			states_e[prev_state_start+i] += e*f_b_params[p];
+			lstm_params_e[p--] += e*states_e[prev_state_start+i];
+		}
+		lstm_params_e[p--]+=e;		
+		s1--;		
+	}
+	
+	//cgate
+	for(s = c_state_size-1; s >= 0; s--){
+		states_e[sct+s]+=states_e[s1]*states[si+s];
+		states_e[si+s]+=states_e[s1]*states[sct+s];
+		states_e[prev_cell_start+s]+=states_e[s1]*states[sf+s];
+		states_e[sf+s]+=states_e[s1]*states[prev_cell_start+s];
+		s1--;
+	}
+	
+	//c + tanh
+	for(s = c_state_size-1; s >= 0; s--){
+		e = (1-states[s1]*states[s1])*states_e[s1];
+		for(i = c_proj_size-1; i >= 0; i--){
+			chars_e[char_start+i] += e*f_b_params[p];
+			lstm_params_e[p--] += e*chars[char_start+i];
+		}		
+		for(i = c_state_size-1; i >= 0; i--){
+			states_e[prev_state_start+i]+=e*f_b_params[p];
+			lstm_params_e[p--] +=e*states[prev_state_start+i];
+		}
+		lstm_params_e[p--]+=e;
+		s1--;		
+	}
+	
+	
+	//fgate
+	for(s = c_state_size-1; s >= 0; s--){		
+		e = states[s1]*(1-states[s1])*states_e[s1];
+		for(i = c_proj_size-1; i >= 0; i--){
+			chars_e[char_start+i] += e*f_b_params[p];
+			lstm_params_e[p--] += e*chars[char_start+i];
+		}
+		for(i = c_cell_size-1; i >= 0; i--){
+			states_e[prev_cell_start+i]+=e*f_b_params[p];
+			lstm_params_e[p--] +=e*states[prev_cell_start+i];
+		}
+		for(i = c_state_size-1; i >= 0; i--){
+			states_e[prev_state_start+i]+=e*f_b_params[p];
+			lstm_params_e[p--] +=e*states[prev_state_start+i];
+		}
+		lstm_params_e[p--]+=e;
+		s1--;
+	}
+	
+	//igate
+	for(s = c_state_size-1; s >= 0; s--){		
+		e = states[s1]*(1-states[s1])*states_e[s1];
+		for(i = c_proj_size-1; i >= 0; i--){
+			chars_e[char_start+i] += e*f_b_params[p];
+			lstm_params_e[p--] += e*chars[char_start+i];
+		}
+		for(i = c_cell_size-1; i >= 0; i--){
+			states_e[prev_cell_start+i]+=e*f_b_params[p];
+			lstm_params_e[p--] +=e*states[prev_cell_start+i];
+		}
+		for(i = c_state_size-1; i >= 0; i--){
+			states_e[prev_state_start+i]+=e*f_b_params[p];
+			lstm_params_e[p--] +=e*states[prev_state_start+i];
+		}
+		lstm_params_e[p--]+=e;
+		s1--;
+	}
+	
+	if(p+1!=pStart){
+		printf("crap! p!= %d p = %d\n",pStart,p+1);
+	}
+	if(s1+1!=next_start){
+		printf("crap! s1!= %d s1 = %d\n",next_start,s1+1);
+	}
+}
+
+void lstmForward(char* word, int len, real* out, real *f_states, real *b_states, real *chars){
+	//printf("%s\n",word);
+	int i,s,c,p;
+	for(s = 0; s < (len+1)*(c_state_size*7); s++){
+		f_states[s]=0;
+		b_states[s]=0;
+	}
+	for(s = 0; s < c_state_size; s++){
+		f_states[c_state_size*3]=f_init_cell[s];
+		f_states[c_state_size*6]=f_init_state[s];
+		b_states[c_state_size*3]=b_init_cell[s];
+		b_states[c_state_size*6]=b_init_state[s];
+	}
+	for(i = 0; i < len; i++){
+		c = word[i];
+		if(c>=C_MAX_CODE){c=C_MAX_CODE-1;}
+		for(s = 0; s < c_proj_size; s++){
+			chars[i*c_proj_size+s] = c_lookup[c*c_proj_size+s];
+		}
+	}
+	
+	for(i = 0; i < len; i++){
+		lstmForwardBlock(chars, i*c_proj_size, f_states, (i+1)*c_state_size*7, 0);
+	}
+	for(i = 0; i < len; i++){
+		lstmForwardBlock(chars, (len-i-1)*c_proj_size, b_states, (i+1)*c_state_size*7, c_lstm_params_number);
+	}
+	
+	//printStates(f_states,c_state_size*7);
+
+	for(s = 0; s < layer1_size; s++){
+		out[s]=0;
+	}	
+	p=c_lstm_params_number*2;
+	for(s = 0; s < layer1_size; s++){
+		for(i = 0; i < c_state_size; i++){
+			out[s]+=f_states[len*c_state_size*7+c_state_size*6 + i]*f_b_params[p++];			
+			out[s]+=b_states[len*c_state_size*7+c_state_size*6 + i]*f_b_params[p++];			
+		}
+//		printf("%f ",out[s]);
+	}
+//	printf("\n");
+}
+
+void lstmBackward(char* word, int len, real* out, real *f_states, real *b_states, real* chars, real* out_e, real *f_states_e, real *b_states_e, real* chars_e, real *lstm_params_e){
+	int i,s,c=-1,p;
+	for(s = 0; s < (len+1)*c_state_size*7; s++){
+		f_states_e[s]=0;
+		b_states_e[s]=0;
+	}
+	for(i = 0; i < len; i++){
+		for(s = 0; s < c_proj_size; s++){
+			chars_e[i*c_proj_size+s] = 0;
+		}
+	}
+	for(i = 0; i < c_lstm_params_number*2; i++){
+		lstm_params_e[i]=0;
+	}
+	
+	p=c_lstm_params_number*2;
+	for(s = 0; s < layer1_size; s++){
+		for(i = 0; i < c_state_size; i++){
+			f_states_e[len*c_state_size*7+c_state_size*6 + i]+=out_e[s]*f_b_params[p];
+			f_b_params[p] += out_e[s] * f_states[len*c_state_size*7+c_state_size*6 + i];
+			p++;
+			b_states_e[len*c_state_size*7+c_state_size*6 + i]+=out_e[s]*f_b_params[p];
+			f_b_params[p] += out_e[s] * b_states[len*c_state_size*7+c_state_size*6 + i];
+			p++;
+		}
+	}
+	
+	for(i = len-1; i >=0; i--){
+		lstmBackwardBlock(chars, i*c_proj_size, b_states, (i+1)*c_state_size*7, c_lstm_params_number, chars_e,b_states_e,lstm_params_e);
+	}
+	
+	for(i = len-1; i >=0; i--){
+		lstmBackwardBlock(chars, (len-i-1)*c_proj_size, f_states, (i+1)*c_state_size*7, 0, chars_e,f_states_e,lstm_params_e);
+	}
+	
+	for(i = 0; i < len; i++){
+		c = word[i];
+		if(c>=C_MAX_CODE){c=C_MAX_CODE-1;}
+		for(s = 0; s < c_proj_size; s++){
+			c_lookup[c*c_proj_size+s] += chars_e[i*c_proj_size+s];
+		}
+	}
+	
+	for(s = 0; s < c_state_size; s++){
+		f_init_cell[s]+=f_states_e[c_state_size*3];
+		f_init_state[s]+=f_states_e[c_state_size*6];
+		b_init_cell[s]+=b_states_e[c_state_size*3];
+		b_init_state[s]+=b_states_e[c_state_size*6];
+	}
+	
+	for(s = 0; s < c_lstm_params_number*2; s++){
+		f_b_params[c]+=lstm_params_e[c];
+	}
+		
+	//printf("out\n");
+	//printStates(f_states,(len)*c_state_size*7);
+	//printf("err\n");
+	//printStates(f_states_e,(len)*c_state_size*7);
+
+}
+
+void lstmFitting(char* word, int len, real* out, real *f_states, real *b_states, real* chars, real* out_expected, real* out_e, real *f_states_e, real *b_states_e, real* chars_e, real *lstm_params_e){
+	int i;
+	real g = 0;
+	lstmForward(word, len, out, f_states, b_states, chars);	
+	for(i = 0; i < layer1_size; i++){
+		if(out_expected[i]>out[i]){
+			g += out_expected[i]-out[i];		
+		}
+		else{
+			g += -out_expected[i]+out[i];		
+		}
+		out_e[i] = (out_expected[i]-out[i])*alpha;
+	}
+	printf("error before fitting = %f\n", g);
+	lstmBackward(word, len, out, f_states, b_states, chars, out_e, f_states_e, b_states_e, chars_e, lstm_params_e);
+	lstmForward(word, len, out, f_states, b_states, chars);	
+	g=0;
+	for(i = 0; i < layer1_size; i++){
+		if(out_expected[i]>out[i]){
+			g += out_expected[i]-out[i];		
+		}
+		else{
+			g += -out_expected[i]+out[i];		
+		}
+		out_e[i] = (out_expected[i]-out[i])*alpha;
+	}
+	printf("error after fitting = %f\n", g);
+	
+}
+
+real hardTanh(real x){
+	if(x>=1){
+		return 1;
+	}
+	else if(x<=-1){
+		return -1;
+	}
+	else{
+		return x;
+	}
+}
+
+real dHardTanh(real x, real g){
+	if(x > 1 && g > 0){
+		return 0;
+	}
+	if(x < -1 && g < 0){
+		return 0;
+	}
+	return 1;
+}
+
+void InitUnigramTable() {
+  int a, i;
+  long long train_words_pow = 0;
+  real d1, power = 0.75;
+  table = (int *)malloc(table_size * sizeof(int));
+  for (a = 0; a < vocab_size; a++) train_words_pow += pow(vocab[a].cn, power);
+  i = 0;
+  d1 = pow(vocab[i].cn, power) / (real)train_words_pow;
+  for (a = 0; a < table_size; a++) {
+    table[a] = i;
+    if (a / (real)table_size > d1) {
+      i++;
+      d1 += pow(vocab[i].cn, power) / (real)train_words_pow;
+    }
+    if (i >= vocab_size) i = vocab_size - 1;
+  }
+}
+
+// Reads a single word from a file, assuming space + tab + EOL to be word boundaries
+void ReadWord(char *word, FILE *fin) {
+  int a = 0, ch;
+  while (!feof(fin)) {
+    ch = fgetc(fin);
+    if (ch == 13) continue;
+    if ((ch == ' ') || (ch == '\t') || (ch == '\n')) {
+      if (a > 0) {
+        if (ch == '\n') ungetc(ch, fin);
+        break;
+      }
+      if (ch == '\n') {
+        strcpy(word, (char *)"</s>");
+        return;
+      } else continue;
+    }
+    word[a] = ch;
+    a++;
+    if (a >= MAX_STRING - 1) a--;   // Truncate too long words
+  }
+  word[a] = 0;
+}
+
+// Returns hash value of a word
+int GetWordHash(char *word) {
+  unsigned long long a, hash = 0;
+  for (a = 0; a < strlen(word); a++) hash = hash * 257 + word[a];
+  hash = hash % vocab_hash_size;
+  return hash;
+}
+
+// Returns position of a word in the vocabulary; if the word is not found, returns -1
+int SearchVocab(char *word) {
+  unsigned int hash = GetWordHash(word);
+  while (1) {
+    if (vocab_hash[hash] == -1) return -1;
+    if (!strcmp(word, vocab[vocab_hash[hash]].word)) return vocab_hash[hash];
+    hash = (hash + 1) % vocab_hash_size;
+  }
+  return -1;
+}
+
+// Reads a word and returns its index in the vocabulary
+int ReadWordIndex(FILE *fin) {
+  char word[MAX_STRING];
+  ReadWord(word, fin);
+  if (feof(fin)) return -1;
+  return SearchVocab(word);
+}
+
+// Reads a word and returns its index in the vocabulary
+int ReadAndStoreWordIndex(FILE *fin, char* word) {
+  ReadWord(word, fin);
+  if (feof(fin)) return -1;
+  return SearchVocab(word);
+}
+
+// Adds a word to the vocabulary
+int AddWordToVocab(char *word) {
+  unsigned int hash, length = strlen(word) + 1;
+  if (length > MAX_STRING) length = MAX_STRING;
+  vocab[vocab_size].word = (char *)calloc(length, sizeof(char));
+  strcpy(vocab[vocab_size].word, word);
+  vocab[vocab_size].cn = 0;
+  vocab_size++;
+  // Reallocate memory if needed
+  if (vocab_size + 2 >= vocab_max_size) {
+    vocab_max_size += 1000;
+    vocab = (struct vocab_word *)realloc(vocab, vocab_max_size * sizeof(struct vocab_word));
+  }
+  hash = GetWordHash(word);
+  while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+  vocab_hash[hash] = vocab_size - 1;
+  return vocab_size - 1;
+}
+
+// Used later for sorting by word counts
+int VocabCompare(const void *a, const void *b) {
+    return ((struct vocab_word *)b)->cn - ((struct vocab_word *)a)->cn;
+}
+
+// Sorts the vocabulary by frequency using word counts
+void SortVocab() {
+  int a, size;
+  unsigned int hash;
+  // Sort the vocabulary and keep </s> at the first position
+  qsort(&vocab[1], vocab_size - 1, sizeof(struct vocab_word), VocabCompare);
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  size = vocab_size;
+  train_words = 0;
+  for (a = 0; a < size; a++) {
+    // Words occuring less than min_count times will be discarded from the vocab
+    if ((vocab[a].cn < min_count) && (a != 0)) {
+      vocab_size--;
+      free(vocab[a].word);
+    } else {
+      // Hash will be re-computed, as after the sorting it is not actual
+      hash=GetWordHash(vocab[a].word);
+      while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+      vocab_hash[hash] = a;
+      train_words += vocab[a].cn;
+    }
+  }
+  vocab = (struct vocab_word *)realloc(vocab, (vocab_size + 1) * sizeof(struct vocab_word));
+  // Allocate memory for the binary tree construction
+  for (a = 0; a < vocab_size; a++) {
+    vocab[a].code = (char *)calloc(MAX_CODE_LENGTH, sizeof(char));
+    vocab[a].point = (int *)calloc(MAX_CODE_LENGTH, sizeof(int));
+  }
+}
+
+// Reduces the vocabulary by removing infrequent tokens
+void ReduceVocab() {
+  int a, b = 0;
+  unsigned int hash;
+  for (a = 0; a < vocab_size; a++) if (vocab[a].cn > min_reduce) {
+    vocab[b].cn = vocab[a].cn;
+    vocab[b].word = vocab[a].word;
+    b++;
+  } else free(vocab[a].word);
+  vocab_size = b;
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  for (a = 0; a < vocab_size; a++) {
+    // Hash will be re-computed, as it is not actual
+    hash = GetWordHash(vocab[a].word);
+    while (vocab_hash[hash] != -1) hash = (hash + 1) % vocab_hash_size;
+    vocab_hash[hash] = a;
+  }
+  fflush(stdout);
+  min_reduce++;
+}
+
+// Create binary Huffman tree using the word counts
+// Frequent words will have short uniqe binary codes
+void CreateBinaryTree() {
+  long long a, b, i, min1i, min2i, pos1, pos2, point[MAX_CODE_LENGTH];
+  char code[MAX_CODE_LENGTH];
+  long long *count = (long long *)calloc(vocab_size * 2 + 1, sizeof(long long));
+  long long *binary = (long long *)calloc(vocab_size * 2 + 1, sizeof(long long));
+  long long *parent_node = (long long *)calloc(vocab_size * 2 + 1, sizeof(long long));
+  for (a = 0; a < vocab_size; a++) count[a] = vocab[a].cn;
+  for (a = vocab_size; a < vocab_size * 2; a++) count[a] = 1e15;
+  pos1 = vocab_size - 1;
+  pos2 = vocab_size;
+  // Following algorithm constructs the Huffman tree by adding one node at a time
+  for (a = 0; a < vocab_size - 1; a++) {
+    // First, find two smallest nodes 'min1, min2'
+    if (pos1 >= 0) {
+      if (count[pos1] < count[pos2]) {
+        min1i = pos1;
+        pos1--;
+      } else {
+        min1i = pos2;
+        pos2++;
+      }
+    } else {
+      min1i = pos2;
+      pos2++;
+    }
+    if (pos1 >= 0) {
+      if (count[pos1] < count[pos2]) {
+        min2i = pos1;
+        pos1--;
+      } else {
+        min2i = pos2;
+        pos2++;
+      }
+    } else {
+      min2i = pos2;
+      pos2++;
+    }
+    count[vocab_size + a] = count[min1i] + count[min2i];
+    parent_node[min1i] = vocab_size + a;
+    parent_node[min2i] = vocab_size + a;
+    binary[min2i] = 1;
+  }
+  // Now assign binary code to each vocabulary word
+  for (a = 0; a < vocab_size; a++) {
+    b = a;
+    i = 0;
+    while (1) {
+      code[i] = binary[b];
+      point[i] = b;
+      i++;
+      b = parent_node[b];
+      if (b == vocab_size * 2 - 2) break;
+    }
+    vocab[a].codelen = i;
+    vocab[a].point[0] = vocab_size - 2;
+    for (b = 0; b < i; b++) {
+      vocab[a].code[i - b - 1] = code[b];
+      vocab[a].point[i - b] = point[b] - vocab_size;
+    }
+  }
+  free(count);
+  free(binary);
+  free(parent_node);
+}
+
+void LearnVocabFromTrainFile() {
+  char word[MAX_STRING];
+  FILE *fin;
+  long long a, i;
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  fin = fopen(train_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: training data file not found!\n");
+    exit(1);
+  }
+  vocab_size = 0;
+  AddWordToVocab((char *)"</s>");
+  while (1) {
+    ReadWord(word, fin);
+    if (feof(fin)) break;
+    train_words++;
+    if ((debug_mode > 1) && (train_words % 100000 == 0)) {
+      printf("%lldK%c", train_words / 1000, 13);
+      fflush(stdout);
+    }
+    i = SearchVocab(word);
+    if (i == -1) {
+      a = AddWordToVocab(word);
+      vocab[a].cn = 1;
+    } else vocab[i].cn++;
+    if (vocab_size > vocab_hash_size * 0.7) ReduceVocab();
+  }
+  SortVocab();
+  if (debug_mode > 0) {
+    printf("Vocab size: %lld\n", vocab_size);
+    printf("Words in train file: %lld\n", train_words);
+  }
+  file_size = ftell(fin);
+  fclose(fin);
+}
+
+void SaveVocab() {
+  long long i;
+  FILE *fo = fopen(save_vocab_file, "wb");
+  for (i = 0; i < vocab_size; i++) fprintf(fo, "%s %lld\n", vocab[i].word, vocab[i].cn);
+  fclose(fo);
+}
+
+void ReadVocab() {
+  long long a, i = 0;
+  char c;
+  char word[MAX_STRING];
+  FILE *fin = fopen(read_vocab_file, "rb");
+  if (fin == NULL) {
+    printf("Vocabulary file not found\n");
+    exit(1);
+  }
+  for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
+  vocab_size = 0;
+  while (1) {
+    ReadWord(word, fin);
+    if (feof(fin)) break;
+    a = AddWordToVocab(word);
+    fscanf(fin, "%lld%c", &vocab[a].cn, &c);
+    i++;
+  }
+  SortVocab();
+  if (debug_mode > 0) {
+    printf("Vocab size: %lld\n", vocab_size);
+    printf("Words in train file: %lld\n", train_words);
+  }
+  fin = fopen(train_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: training data file not found!\n");
+    exit(1);
+  }
+  fseek(fin, 0, SEEK_END);
+  file_size = ftell(fin);
+  fclose(fin);
+}
+
+void InitClassUnigramTable() {
+  long long a,c;
+  printf("loading class unigrams \n");
+  FILE *fin = fopen(negative_classes_file, "rb");
+  if (fin == NULL) {
+    printf("ERROR: class file not found!\n");
+    exit(1);
+  }
+  word_to_group = (int *)malloc(vocab_size * sizeof(int));
+  for(a = 0; a < vocab_size; a++) word_to_group[a] = -1;
+  char class[MAX_STRING];
+  char prev_class[MAX_STRING];
+  prev_class[0] = 0;
+  char word[MAX_STRING];
+  class_number = -1;
+  while (1) {
+    if (feof(fin)) break;
+    ReadWord(class, fin);
+    ReadWord(word, fin);
+    int word_index = SearchVocab(word);
+    if (word_index != -1){
+       if(strcmp(class, prev_class) != 0){
+	    class_number++;
+	    strcpy(prev_class, class);
+       }
+       word_to_group[word_index] = class_number;
+    }
+    ReadWord(word, fin);
+  }
+  class_number++;
+  fclose(fin);
+  
+  group_to_table = (int *)malloc(table_size * class_number * sizeof(int)); 
+  long long train_words_pow = 0;
+  real d1, power = 0.75;
+  
+  for(c = 0; c < class_number; c++){
+     long long offset = c * table_size;
+     train_words_pow = 0;
+     for (a = 0; a < vocab_size; a++) if(word_to_group[a] == c) train_words_pow += pow(vocab[a].cn, power);
+     int i = 0;
+     while(word_to_group[i]!=c && i < vocab_size) i++;
+     d1 = pow(vocab[i].cn, power) / (real)train_words_pow;
+     for (a = 0; a < table_size; a++) {
+	//printf("index %lld , word %d\n", a, i);
+	group_to_table[offset + a] = i;
+        if (a / (real)table_size > d1) {
+	   i++;
+           while(word_to_group[i]!=c && i < vocab_size) i++;
+	   d1 += pow(vocab[i].cn, power) / (real)train_words_pow;
+        }
+        if (i >= vocab_size) while(word_to_group[i]!=c && i >= 0) i--;
+     }
+  }
+}
+
+void InitNet() {
+  long long a, b;
+  unsigned long long next_random = 1;
+  window_layer_size = layer1_size*window*2;
+  a = posix_memalign((void **)&syn0, 128, (long long)vocab_size * layer1_size * sizeof(real));
+  if (syn0 == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  
+  if (hs) {
+    a = posix_memalign((void **)&syn1, 128, (long long)vocab_size * layer1_size * sizeof(real));
+    if (syn1 == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn1_window, 128, (long long)vocab_size * window_layer_size * sizeof(real));
+    if (syn1_window == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn_hidden_word, 128, (long long)vocab_size * window_hidden_size * sizeof(real));
+    if (syn_hidden_word == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++)
+     syn1[a * layer1_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_layer_size; b++)
+     syn1_window[a * window_layer_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_hidden_size; b++)
+     syn_hidden_word[a * window_hidden_size + b] = 0;
+  }
+  if (negative>0) {
+    a = posix_memalign((void **)&syn1neg, 128, (long long)vocab_size * layer1_size * sizeof(real));
+    if (syn1neg == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn1neg_window, 128, (long long)vocab_size * window_layer_size * sizeof(real));
+    if (syn1neg_window == NULL) {printf("Memory allocation failed\n"); exit(1);}
+    a = posix_memalign((void **)&syn_hidden_word_neg, 128, (long long)vocab_size * window_hidden_size * sizeof(real));
+    if (syn_hidden_word_neg == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++)
+     syn1neg[a * layer1_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_layer_size; b++)
+     syn1neg_window[a * window_layer_size + b] = 0;
+    for (a = 0; a < vocab_size; a++) for (b = 0; b < window_hidden_size; b++)
+     syn_hidden_word_neg[a * window_hidden_size + b] = 0;
+  }
+  for (a = 0; a < vocab_size; a++) for (b = 0; b < layer1_size; b++) {
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    syn0[a * layer1_size + b] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / layer1_size;
+  }
+
+  a = posix_memalign((void **)&syn_window_hidden, 128, window_hidden_size * window_layer_size * sizeof(real));
+  if (syn_window_hidden == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  for (a = 0; a < window_hidden_size * window_layer_size; a++){
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    syn_window_hidden[a] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / (window_hidden_size*window_layer_size);
+  }
+  
+  if(rep == 1 || rep == 2){
+  a = posix_memalign((void **)&c_lookup, 128, (long long)C_MAX_CODE * c_proj_size * sizeof(real));
+  if (c_lookup == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  for (a = 0; a < C_MAX_CODE * c_proj_size; a++){
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    c_lookup[a] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / (c_proj_size);
+  }
+
+  a = posix_memalign((void **)&f_init_state, 128, c_state_size * sizeof(real));
+  if (f_init_state == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  a = posix_memalign((void **)&f_init_cell, 128, c_state_size * sizeof(real));
+  if (f_init_cell == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  a = posix_memalign((void **)&b_init_state, 128, c_state_size * sizeof(real));
+  if (b_init_state == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  a = posix_memalign((void **)&b_init_cell, 128, c_state_size * sizeof(real));
+  if (b_init_cell == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+  for (a = 0; a < c_state_size; a++){
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    f_init_state[a] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / (c_state_size);
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    f_init_cell[a] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / (c_state_size);
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    b_init_state[a] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / (c_state_size);
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    b_init_cell[a] = (((next_random & 0xFFFF) / (real)65536) - 0.5) / (c_state_size);
+  }
+
+  c_lstm_params_number = /*input*/ (c_state_size+c_cell_size+c_proj_size+1)*c_state_size +
+  /*forget*/ (c_state_size+c_cell_size+c_proj_size+1)*c_state_size +
+  /*cell*/ (c_state_size+c_proj_size+1)*c_state_size +
+  /*output*/ (c_state_size+c_cell_size+c_proj_size+1)*c_state_size;
+  
+  c_params_number = ( c_lstm_params_number * 2 + (c_state_size*2)*layer1_size) ;
+  a = posix_memalign((void **)&f_b_params, 128, c_params_number* sizeof(real));
+  if (f_b_params == NULL) {printf("Memory allocation failed\n"); exit(1);}
+
+  for (a = 0; a < c_params_number; a++){
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    f_b_params[a] = (((next_random & 0xFFFF) / (real)65536) - 0.5) ;
+  }
+  }
+  
+  if(rep == 2){
+  	  a = posix_memalign((void **)&syn0_initial, 128, (long long)vocab_size * layer1_size * sizeof(real));
+    if (syn0_initial == NULL) {printf("Memory allocation failed\n"); exit(1);}
+  	  a = posix_memalign((void **)&syn0_in_memory, 128, (long long)vocab_size * sizeof(real));
+	if (syn0_in_memory == NULL) {printf("Memory allocation failed\n"); exit(1);}
+	for(a = 0; a < vocab_size; a++){
+		syn0_in_memory[a] = -1;
+	}
+  }
+  CreateBinaryTree();
+}
+
+void *TrainModelThread(void *id) {
+  long long a, b, d, cw, word, last_word, sentence_length = 0, sentence_position = 0;
+  long long word_count = 0, last_word_count = 0, sen[MAX_SENTENCE_LENGTH + 1];
+  long long l1, l2, c, target, label, local_iter = iter;
+  char c_sen[(MAX_SENTENCE_LENGTH + 1) * MAX_STRING];
+  unsigned long long next_random = (long long)id;
+  real f, g, acc_g=0;
+  clock_t now;
+  int input_len_1 = layer1_size;
+  if(type == 2 || type == 4){
+     input_len_1=window_layer_size;
+  }
+  real *neu1 = (real *)calloc(input_len_1, sizeof(real));
+  real *neu1e = (real *)calloc(input_len_1, sizeof(real));
+
+  int input_len_2 = 0;
+  if(type == 4){
+     input_len_2 = window_hidden_size;
+  }
+  real *neu2 = (real *)calloc(input_len_2, sizeof(real));
+  real *neu2e = (real *)calloc(input_len_2, sizeof(real));
+
+  FILE *fi = fopen(train_file, "rb");
+  fseek(fi, file_size / (long long)num_threads * (long long)id, SEEK_SET);
+  
+  real *f_states = (real *)calloc((c_state_size * 7) * (MAX_STRING + 1), sizeof(real));
+  real *f_states_e = (real *)calloc((c_state_size * 7) * (MAX_STRING + 1), sizeof(real));
+  real *b_states = (real *)calloc((c_state_size * 7) * (MAX_STRING + 1), sizeof(real));
+  real *b_states_e = (real *)calloc((c_state_size * 7) * (MAX_STRING + 1), sizeof(real));
+  real *chars = (real *)calloc(c_proj_size * MAX_STRING, sizeof(real));
+  real *chars_e = (real *)calloc(c_proj_size * MAX_STRING, sizeof(real));
+  real *lstm_params_e = (real *)calloc(c_lstm_params_number*2, sizeof(real));
+  
+  //short term memory vars
+  real global_divergence = -1;
+  int in_mem = 0;
+  int skip=0, non_skip=0;
+  
+  while (1) {
+    if (word_count - last_word_count > 10000) {
+      word_count_actual += word_count - last_word_count;
+      last_word_count = word_count;
+      if ((debug_mode > 1)) {
+        now=clock();
+        printf("%cAlpha: %f  Progress: %.2f%%  Words/thread/sec: %.2fk : error %.4f", 13, alpha,
+         word_count_actual / (real)(iter * train_words + 1) * 100,
+         word_count_actual / ((real)(now - start + 1) / (real)CLOCKS_PER_SEC * 1000), acc_g);
+         if(rep == 2){
+         	printf(" skiprate %f",skip/(real)(skip+non_skip));
+         }
+         acc_g=0;
+         skip=0;
+         non_skip=0;
+        fflush(stdout);
+      }
+      alpha = starting_alpha * (1 - word_count_actual / (real)(iter * train_words + 1));
+      if (alpha < starting_alpha * 0.0001) alpha = starting_alpha * 0.0001;
+    }
+    if (sentence_length == 0) {
+      while (1) {
+        word = ReadAndStoreWordIndex(fi, &c_sen[sentence_length*MAX_STRING]);
+        if (feof(fi)) break;
+        if (word == -1) continue;
+        word_count++;
+        if (word == 0) break;
+        // The subsampling randomly discards frequent words while keeping the ranking same
+        if (sample > 0) {
+          real ran = (sqrt(vocab[word].cn / (sample * train_words)) + 1) * (sample * train_words) / vocab[word].cn;
+          next_random = next_random * (unsigned long long)25214903917 + 11;
+          if (ran < (next_random & 0xFFFF) / (real)65536) continue;
+        }
+        sen[sentence_length] = word;
+        sentence_length++;
+        if (sentence_length >= MAX_SENTENCE_LENGTH) break;
+      }
+      sentence_position = 0;
+    }
+    if (feof(fi) || (word_count > train_words / num_threads)) {
+      word_count_actual += word_count - last_word_count;
+      local_iter--;
+      if (local_iter == 0) break;
+      word_count = 0;
+      last_word_count = 0;
+      sentence_length = 0;
+      fseek(fi, file_size / (long long)num_threads * (long long)id, SEEK_SET);
+      continue;
+    }
+    word = sen[sentence_position];
+    if (word == -1) continue;
+    for (c = 0; c < input_len_1; c++) neu1[c] = 0;
+    for (c = 0; c < input_len_1; c++) neu1e[c] = 0;
+    for (c = 0; c < input_len_2; c++) neu2[c] = 0;
+    for (c = 0; c < input_len_2; c++) neu2e[c] = 0;
+    next_random = next_random * (unsigned long long)25214903917 + 11;
+    b = next_random % window;
+    if (type == 0) {  //train the cbow architecture
+      // in -> hidden
+      cw = 0;
+      for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        for (c = 0; c < layer1_size; c++) neu1[c] += syn0[c + last_word * layer1_size];
+        cw++;
+      }
+      if (cw) {
+        for (c = 0; c < layer1_size; c++) neu1[c] /= cw;
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * layer1_size;
+          // Propagate hidden -> output
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < layer1_size; c++) syn1[c + l2] += g * neu1[c];
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+	    if(word_to_group != NULL && word_to_group[word] != -1){
+		target = word;
+		while(target == word) {
+			target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+            		next_random = next_random * (unsigned long long)25214903917 + 11;
+		}
+		//printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+	    }
+	    else{
+            	target = table[(next_random >> 16) % table_size];
+	    }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1neg[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha;
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1neg[c + l2] += g * neu1[c];
+        }
+        // hidden -> in
+        for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {
+          c = sentence_position - window + a;
+          if (c < 0) continue;
+          if (c >= sentence_length) continue;
+          last_word = sen[c];
+          if (last_word == -1) continue;
+          for (c = 0; c < layer1_size; c++) syn0[c + last_word * layer1_size] += neu1e[c];
+        }
+      }
+    } else if(type==1) {  //train skip-gram
+      for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        l1 = last_word * layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1e[c] = 0;
+        // HIERARCHICAL SOFTMAX
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * layer1_size;
+          // Propagate hidden -> output
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < layer1_size; c++) syn1[c + l2] += g * syn0[c + l1];
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	    next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * layer1_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += syn0[c + l1] * syn1neg[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha;
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg[c + l2];
+          for (c = 0; c < layer1_size; c++) syn1neg[c + l2] += g * syn0[c + l1];
+        }
+        // Learn weights input -> hidden
+        for (c = 0; c < layer1_size; c++) syn0[c + l1] += neu1e[c];
+      }
+    }
+    else if(type == 2){ //train the cwindow architecture
+      // in -> hidden
+      cw = 0;
+      for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        window_offset = a*layer1_size;
+        if (a > window) window_offset-=layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1[c+window_offset] += syn0[c + last_word * layer1_size];
+        cw++;
+      }
+      if (cw) {
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * window_layer_size;
+          // Propagate hidden -> output
+          for (c = 0; c < window_layer_size; c++) f += neu1[c] * syn1_window[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < window_layer_size; c++) neu1e[c] += g * syn1_window[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < window_layer_size; c++) syn1_window[c + l2] += g * neu1[c];
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < window_layer_size; c++) f += neu1[c] * syn1neg_window[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha;
+          acc_g+=g;
+          for (c = 0; c < window_layer_size; c++) neu1e[c] += g * syn1neg_window[c + l2];
+          for (c = 0; c < window_layer_size; c++) syn1neg_window[c + l2] += g * neu1[c];
+        }
+        // hidden -> in
+        for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+          c = sentence_position - window + a;
+          if (c < 0) continue;
+          if (c >= sentence_length) continue;
+          last_word = sen[c];
+          if (last_word == -1) continue;
+	  window_offset = a * layer1_size;
+	  if(a > window) window_offset -= layer1_size;
+          for (c = 0; c < layer1_size; c++) syn0[c + last_word * layer1_size] += neu1e[c + window_offset];
+        }
+      }
+    }
+    else if (type == 3){  //train structured skip-gram
+       char* c_word = &c_sen[sentence_position*MAX_STRING];
+       if(rep == 1){
+        	lstmForward(c_word, strlen(c_word),neu1, f_states, b_states, chars);
+        } 
+        else if(rep == 2){
+	        l1 = word * layer1_size;
+            if(syn0_in_memory[word]==-1){
+            	syn0_in_memory[word]=0;
+        		lstmForward(c_word, strlen(c_word),&syn0_initial[l1], f_states, b_states, chars);
+	        	for (c = 0; c < layer1_size; c++) {syn0[c + l1] = syn0_initial[c + l1];neu1[c] += syn0[c + l1];}
+	        	in_mem = 1;
+        	}
+        	else{
+	        	for (c = 0; c < layer1_size; c++) neu1[c] += syn0[c + l1];
+	        	in_mem = 0;	        	
+        	}
+        }
+        else{
+        	l1 = word * layer1_size;
+	        for (c = 0; c < layer1_size; c++) neu1[c] += syn0[c + l1];
+        }
+        
+      for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+
+        
+	window_offset = a * layer1_size;
+	if(a > window) window_offset -= layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1e[c] = 0;
+        // HIERARCHICAL SOFTMAX
+        if (hs) for (d = 0; d < vocab[last_word].codelen; d++) {
+          f = 0;
+          l2 = vocab[last_word].point[d] * window_layer_size;
+          // Propagate hidden -> output
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1_window[c + l2 + window_offset];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[last_word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1_window[c + l2 + window_offset];
+          // Learn weights hidden -> output
+          for (c = 0; c < layer1_size; c++) syn1[c + l2 + window_offset] += g * neu1[c];
+        }
+        // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = last_word;
+            label = 1;
+          } else {
+	     next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[last_word] != -1){
+                target = last_word;
+                while(target == last_word) {
+                        target = group_to_table[word_to_group[last_word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == last_word) continue;
+            label = 0;
+          }
+          l2 = target * window_layer_size;
+          f = 0;
+          for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1neg_window[c + l2 + window_offset];
+          if (f > MAX_EXP) g = (label - 1) * alpha;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha;
+          acc_g+=g;
+
+          for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg_window[c + l2 + window_offset];
+          for (c = 0; c < layer1_size; c++) syn1neg_window[c + l2 + window_offset] += g * neu1[c];
+          
+        }
+        
+      }
+      // Learn weights input -> hidden
+        
+        if(rep == 1){
+        	lstmBackward(c_word, strlen(c_word),neu1, f_states, b_states, chars, neu1e,f_states_e, b_states_e, chars_e, lstm_params_e);
+        }
+        else if(rep == 2){
+        	g = 0;
+	        l1 = word * layer1_size;
+        	for (c = 0; c < layer1_size; c++) {
+        		syn0[c + l1] += neu1e[c];
+        		f = syn0[c + l1] - syn0_initial[c + l1];
+        		if(f > 0){
+        			g+=f;
+        		}
+        		else{
+        			g-=f;
+        		}
+        	}
+        	syn0_in_memory[word] = g;
+        	if(global_divergence == -1){global_divergence = g;}
+        	long skip_prob = vocab[word].cn-(log(vocab[word].cn)+1);
+            next_random = next_random * (unsigned long long)25214903917 + 11;
+			
+        	if(skip_prob < next_random%vocab[word].cn){
+        		non_skip++;
+        		if(in_mem == 0){
+        			lstmFitting(c_word, strlen(c_word),neu1, f_states, b_states, chars,&syn0[c +l1], neu1e,f_states_e, b_states_e, chars_e, lstm_params_e);
+        		}
+        		else{    		
+	        		lstmBackward(c_word, strlen(c_word),neu1, f_states, b_states, chars, neu1e,f_states_e, b_states_e, chars_e, lstm_params_e);	        	
+	        	}
+            	syn0_in_memory[word]=-1;
+        	}
+        	else{
+        		skip++;
+        	}
+       		global_divergence = global_divergence*0.9 + g*0.1;
+        }
+        else{
+   	        l1 = word * layer1_size;
+        	for (c = 0; c < layer1_size; c++) syn0[c + l1] += neu1e[c];
+        }
+    }
+    else if(type == 4){ //training senna
+	// in -> hidden
+      cw = 0;
+      for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+        c = sentence_position - window + a;
+        if (c < 0) continue;
+        if (c >= sentence_length) continue;
+        last_word = sen[c];
+        if (last_word == -1) continue;
+        window_offset = a*layer1_size;
+        if (a > window) window_offset-=layer1_size;
+        for (c = 0; c < layer1_size; c++) neu1[c+window_offset] += syn0[c + last_word * layer1_size];
+        cw++;
+      }
+      if (cw) {
+		for (a = 0; a < window_hidden_size; a++){
+          c = a*window_layer_size;
+          for(b = 0; b < window_layer_size; b++){
+             neu2[a] += syn_window_hidden[c + b] * neu1[b];
+          }
+        }
+        if (hs) for (d = 0; d < vocab[word].codelen; d++) {
+          f = 0;
+          l2 = vocab[word].point[d] * window_hidden_size;
+          // Propagate hidden -> output
+          for (c = 0; c < window_hidden_size; c++) f += hardTanh(neu2[c]) * syn_hidden_word[c + l2];
+          if (f <= -MAX_EXP) continue;
+          else if (f >= MAX_EXP) continue;
+          else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))];
+          // 'g' is the gradient multiplied by the learning rate
+          g = (1 - vocab[word].code[d] - f) * alpha;
+          // Propagate errors output -> hidden
+          for (c = 0; c < window_hidden_size; c++) neu2e[c] += dHardTanh(neu2[c],g) * g * syn_hidden_word[c + l2];
+          // Learn weights hidden -> output
+          for (c = 0; c < window_hidden_size; c++) syn_hidden_word[c + l2] += dHardTanh(neu2[c],g) * g * neu2[c];
+        }
+      // NEGATIVE SAMPLING
+        if (negative > 0) for (d = 0; d < negative + 1; d++) {
+          if (d == 0) {
+            target = word;
+            label = 1;
+          } else {
+	    next_random = next_random * (unsigned long long)25214903917 + 11;
+            if(word_to_group != NULL && word_to_group[word] != -1){
+                target = word;
+                while(target == word) {
+                        target = group_to_table[word_to_group[word]*table_size + (next_random >> 16) % table_size];
+                        next_random = next_random * (unsigned long long)25214903917 + 11;
+                }
+                //printf("negative sampling %lld for word %s returned %s\n", d, vocab[word].word, vocab[target].word);
+            }
+            else{
+                target = table[(next_random >> 16) % table_size];
+            }
+            if (target == 0) target = next_random % (vocab_size - 1) + 1;
+            if (target == word) continue;
+            label = 0;
+          }
+          l2 = target * window_hidden_size;
+          f = 0;
+          for (c = 0; c < window_hidden_size; c++) f += hardTanh(neu2[c]) * syn_hidden_word_neg[c + l2];
+          if (f > MAX_EXP) g = (label - 1) * alpha / negative;
+          else if (f < -MAX_EXP) g = (label - 0) * alpha / negative;
+          else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha / negative;
+          for (c = 0; c < window_hidden_size; c++) neu2e[c] += dHardTanh(neu2[c],g) * g * syn_hidden_word_neg[c + l2];
+          for (c = 0; c < window_hidden_size; c++) syn_hidden_word_neg[c + l2] += dHardTanh(neu2[c],g) * g * neu2[c];
+        }
+        for (a = 0; a < window_hidden_size; a++)
+          for(b = 0; b < window_layer_size; b++)
+	     neu1e[b] += neu2e[a] * syn_window_hidden[a*window_layer_size + b];
+	for (a = 0; a < window_hidden_size; a++)
+          for(b = 0; b < window_layer_size; b++)
+	     syn_window_hidden[a*window_layer_size + b] += neu2e[a] * neu1[b];
+        // hidden -> in
+        for (a = 0; a < window * 2 + 1; a++) if (a != window) {
+          c = sentence_position - window + a;
+          if (c < 0) continue;
+          if (c >= sentence_length) continue;
+          last_word = sen[c];
+          if (last_word == -1) continue;
+          window_offset = a * layer1_size;
+          if(a > window) window_offset -= layer1_size;
+          for (c = 0; c < layer1_size; c++) syn0[c + last_word * layer1_size] += neu1e[c + window_offset];
+        }
+      }
+    }
+    else{
+	printf("unknown type %i", type);
+	exit(0);
+    }
+    sentence_position++;
+    if (sentence_position >= sentence_length) {
+      sentence_length = 0;
+      continue;
+    }
+  }
+  fclose(fi);
+  free(neu1);
+  free(neu1e);
+  pthread_exit(NULL);
+}
+
+void TrainModel() {
+  long a, b, c, d;
+  FILE *fo;
+  pthread_t *pt = (pthread_t *)malloc(num_threads * sizeof(pthread_t));
+  printf("Starting training using file %s\n", train_file);
+  starting_alpha = alpha;
+  if (read_vocab_file[0] != 0) ReadVocab(); else LearnVocabFromTrainFile();
+  if (save_vocab_file[0] != 0) SaveVocab();
+  if (output_file[0] == 0) return;
+  InitNet();
+  if (negative > 0) InitUnigramTable();
+  if (negative_classes_file[0] != 0) InitClassUnigramTable();
+  start = clock();
+  for (a = 0; a < num_threads; a++) pthread_create(&pt[a], NULL, TrainModelThread, (void *)a);
+  for (a = 0; a < num_threads; a++) pthread_join(pt[a], NULL);
+  fo = fopen(output_file, "wb");
+  if (classes == 0) {
+    // Save the word vectors
+	real *f_states = (real *)calloc((c_state_size * 7) * (MAX_STRING + 1), sizeof(real));
+	real *b_states = (real *)calloc((c_state_size * 7) * (MAX_STRING + 1), sizeof(real));
+  	real *chars = (real *)calloc(c_proj_size * MAX_STRING, sizeof(real));
+	real *neu1 = (real *)calloc(layer1_size * MAX_STRING, sizeof(real));
+
+    fprintf(fo, "%lld %lld\n", vocab_size, layer1_size);
+    for (a = 0; a < vocab_size; a++) {
+      fprintf(fo, "%s ", vocab[a].word);
+      if(rep == 1 || rep == 2){
+      	for (b = 0; b < layer1_size; b++) {neu1[b]=0;}
+      	lstmForward(vocab[a].word, strlen(vocab[a].word),neu1, f_states,b_states,chars);
+        if (binary) for (b = 0; b < layer1_size; b++) fwrite(&neu1[b], sizeof(real), 1, fo);
+        else for (b = 0; b < layer1_size; b++) fprintf(fo, "%lf ", neu1[b]);
+      }
+      else{
+        if (binary) for (b = 0; b < layer1_size; b++) fwrite(&syn0[a * layer1_size + b], sizeof(real), 1, fo);
+        else for (b = 0; b < layer1_size; b++) fprintf(fo, "%lf ", syn0[a * layer1_size + b]);
+      }
+      fprintf(fo, "\n");
+    }
+  } else {
+    // Run K-means on the word vectors
+    int clcn = classes, iter = 10, closeid;
+    int *centcn = (int *)malloc(classes * sizeof(int));
+    int *cl = (int *)calloc(vocab_size, sizeof(int));
+    real closev, x;
+    real *cent = (real *)calloc(classes * layer1_size, sizeof(real));
+    for (a = 0; a < vocab_size; a++) cl[a] = a % clcn;
+    for (a = 0; a < iter; a++) {
+      for (b = 0; b < clcn * layer1_size; b++) cent[b] = 0;
+      for (b = 0; b < clcn; b++) centcn[b] = 1;
+      for (c = 0; c < vocab_size; c++) {
+        for (d = 0; d < layer1_size; d++) cent[layer1_size * cl[c] + d] += syn0[c * layer1_size + d];
+        centcn[cl[c]]++;
+      }
+      for (b = 0; b < clcn; b++) {
+        closev = 0;
+        for (c = 0; c < layer1_size; c++) {
+          cent[layer1_size * b + c] /= centcn[b];
+          closev += cent[layer1_size * b + c] * cent[layer1_size * b + c];
+        }
+        closev = sqrt(closev);
+        for (c = 0; c < layer1_size; c++) cent[layer1_size * b + c] /= closev;
+      }
+      for (c = 0; c < vocab_size; c++) {
+        closev = -10;
+        closeid = 0;
+        for (d = 0; d < clcn; d++) {
+          x = 0;
+          for (b = 0; b < layer1_size; b++) x += cent[layer1_size * d + b] * syn0[c * layer1_size + b];
+          if (x > closev) {
+            closev = x;
+            closeid = d;
+          }
+        }
+        cl[c] = closeid;
+      }
+    }
+    // Save the K-means classes
+    for (a = 0; a < vocab_size; a++) fprintf(fo, "%s %d\n", vocab[a].word, cl[a]);
+    free(centcn);
+    free(cent);
+    free(cl);
+  }
+  fclose(fo);
+}
+
+int ArgPos(char *str, int argc, char **argv) {
+  int a;
+  for (a = 1; a < argc; a++) if (!strcmp(str, argv[a])) {
+    if (a == argc - 1) {
+      printf("Argument missing for %s\n", str);
+      exit(1);
+    }
+    return a;
+  }
+  return -1;
+}
+
+int main(int argc, char **argv) {
+  int i;
+  if (argc == 1) {
+    printf("WORD VECTOR estimation toolkit v 0.1c\n\n");
+    printf("Options:\n");
+    printf("Parameters for training:\n");
+    printf("\t-train <file>\n");
+    printf("\t\tUse text data from <file> to train the model\n");
+    printf("\t-output <file>\n");
+    printf("\t\tUse <file> to save the resulting word vectors / word clusters\n");
+    printf("\t-size <int>\n");
+    printf("\t\tSet size of word vectors; default is 100\n");
+    printf("\t-window <int>\n");
+    printf("\t\tSet max skip length between words; default is 5\n");
+    printf("\t-sample <float>\n");
+    printf("\t\tSet threshold for occurrence of words. Those that appear with higher frequency in the training data\n");
+    printf("\t\twill be randomly down-sampled; default is 1e-3, useful range is (0, 1e-5)\n");
+    printf("\t-hs <int>\n");
+    printf("\t\tUse Hierarchical Softmax; default is 0 (not used)\n");
+    printf("\t-negative <int>\n");
+    printf("\t-negative-classes <file>\n");
+    printf("\t\tNumber of negative examples; default is 5, common values are 3 - 10 (0 = not used)\n");
+    printf("\t-threads <int>\n");
+    printf("\t\tUse <int> threads (default 12)\n");
+    printf("\t-iter <int>\n");
+    printf("\t\tRun more training iterations (default 5)\n");
+    printf("\t-min-count <int>\n");
+    printf("\t\tThis will discard words that appear less than <int> times; default is 5\n");
+    printf("\t-alpha <float>\n");
+    printf("\t\tSet the starting learning rate; default is 0.025 for skip-gram and 0.05 for CBOW\n");
+    printf("\t-classes <int>\n");
+    printf("\t\tOutput word classes rather than word vectors; default number of classes is 0 (vectors are written)\n");
+    printf("\t-debug <int>\n");
+    printf("\t\tSet the debug mode (default = 2 = more info during training)\n");
+    printf("\t-binary <int>\n");
+    printf("\t\tSave the resulting vectors in binary moded; default is 0 (off)\n");
+    printf("\t-save-vocab <file>\n");
+    printf("\t\tThe vocabulary will be saved to <file>\n");
+    printf("\t-read-vocab <file>\n");
+    printf("\t\tThe vocabulary will be read from <file>, not constructed from the training data\n");
+    printf("\t-type <int>\n");
+    printf("\t\tType of embeddings (0 for cbow, 1 for skipngram, 2 for cwindow, 3 for structured skipngram, 4 for senna type)\n");
+    printf("\t-rep <int>\n");
+    printf("\t\tType of word rep (0 for word, 1 for character, 2 for character with short term memory\n");
+    printf("\t-char-state-dim <int>\n");
+    printf("\t\tcharacter state size\n");
+    printf("\t-char-proj-dim <int>\n");
+    printf("\t\tcharacter projection size\n");
+    printf("\nExamples:\n");
+    printf("./word2vec -train data.txt -output vec.txt -size 200 -window 5 -sample 1e-4 -negative 5 -hs 0 -binary 0 -type 1 -iter 3\n\n");
+    return 0;
+  }
+  output_file[0] = 0;
+  save_vocab_file[0] = 0;
+  read_vocab_file[0] = 0;
+  negative_classes_file[0] = 0;
+  if ((i = ArgPos((char *)"-size", argc, argv)) > 0) layer1_size = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-train", argc, argv)) > 0) strcpy(train_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-save-vocab", argc, argv)) > 0) strcpy(save_vocab_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-read-vocab", argc, argv)) > 0) strcpy(read_vocab_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-debug", argc, argv)) > 0) debug_mode = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-binary", argc, argv)) > 0) binary = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-type", argc, argv)) > 0) type = atoi(argv[i + 1]);
+  if (type==0 || type==2 || type==4) alpha = 0.05;
+  if ((i = ArgPos((char *)"-alpha", argc, argv)) > 0) alpha = atof(argv[i + 1]);
+  if ((i = ArgPos((char *)"-output", argc, argv)) > 0) strcpy(output_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-window", argc, argv)) > 0) window = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-sample", argc, argv)) > 0) sample = atof(argv[i + 1]);
+  if ((i = ArgPos((char *)"-hs", argc, argv)) > 0) hs = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-negative", argc, argv)) > 0) negative = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-negative-classes", argc, argv)) > 0) strcpy(negative_classes_file, argv[i + 1]);
+  if ((i = ArgPos((char *)"-threads", argc, argv)) > 0) num_threads = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-iter", argc, argv)) > 0) iter = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-min-count", argc, argv)) > 0) min_count = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-classes", argc, argv)) > 0) classes = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-rep", argc, argv)) > 0) rep = atoi(argv[i + 1]);
+  if ((i = ArgPos((char *)"-char-state-dim", argc, argv)) > 0) {c_state_size = atoi(argv[i + 1]); c_cell_size = c_state_size;}
+  if ((i = ArgPos((char *)"-char-proj-dim", argc, argv)) > 0) {c_proj_size = atoi(argv[i + 1]);}
+  vocab = (struct vocab_word *)calloc(vocab_max_size, sizeof(struct vocab_word));
+  vocab_hash = (int *)calloc(vocab_hash_size, sizeof(int));
+  expTable = (real *)malloc((EXP_TABLE_SIZE + 1) * sizeof(real));
+  tanhTable = (real *)malloc((EXP_TABLE_SIZE + 1) * sizeof(real));
+  for (i = 0; i < EXP_TABLE_SIZE; i++) {
+    expTable[i] = exp((i / (real)EXP_TABLE_SIZE * 2 - 1) * MAX_EXP); // Precompute the exp() table
+    expTable[i] = expTable[i] / (expTable[i] + 1);                   // Precompute f(x) = x / (x + 1)
+    tanhTable[i] = tanh((i / (real)EXP_TABLE_SIZE * 2 - 1) * MAX_EXP);
+  }
+  TrainModel();
+  return 0;
+}