Marc Kupietz | dc22b98 | 2015-10-09 09:19:34 +0200 | [diff] [blame] | 1 | #!/usr/local/bin/perl |
| 2 | use Inline C; |
| 3 | use Mojolicious::Lite; |
Marc Kupietz | 247500f | 2015-10-09 11:29:01 +0200 | [diff] [blame^] | 4 | use Encode qw(decode encode); |
Marc Kupietz | dc22b98 | 2015-10-09 09:19:34 +0200 | [diff] [blame] | 5 | |
Marc Kupietz | 247500f | 2015-10-09 11:29:01 +0200 | [diff] [blame^] | 6 | init_net("vectors14.bin"); |
Marc Kupietz | dc22b98 | 2015-10-09 09:19:34 +0200 | [diff] [blame] | 7 | |
| 8 | get '/' => sub { |
| 9 | my $c = shift; |
| 10 | my $word=$c->param('word'); |
Marc Kupietz | 247500f | 2015-10-09 11:29:01 +0200 | [diff] [blame^] | 11 | my $list; |
| 12 | if($word !~ /^\s*$/) { |
| 13 | $list = print_neighbours(encode("iso-8859-1", $word)); |
| 14 | } |
| 15 | $c->render(template=>"index", word=>$word, list=> $list); |
Marc Kupietz | dc22b98 | 2015-10-09 09:19:34 +0200 | [diff] [blame] | 16 | }; |
| 17 | |
| 18 | app->start; |
| 19 | |
| 20 | exit; |
| 21 | |
| 22 | __END__ |
| 23 | |
| 24 | __C__ |
| 25 | #include <stdio.h> |
| 26 | #include <string.h> |
| 27 | #include <math.h> |
| 28 | #include <malloc.h> |
| 29 | #include <stdlib.h> //strlen |
| 30 | |
| 31 | #define max_size 2000 |
| 32 | #define max_w 50 |
| 33 | #define N 75 |
| 34 | |
| 35 | //the thread function |
| 36 | void *connection_handler(void *); |
| 37 | |
| 38 | char *bestw[N]; |
| 39 | char file_name[max_size], st[100][max_size]; |
| 40 | float dist, len, bestd[N], vec[max_size]; |
| 41 | long long words, size, a, b, c, d, cn, bi[100]; |
| 42 | char ch; |
| 43 | float *M; |
| 44 | char *vocab; |
| 45 | char *stringBuffer; |
| 46 | |
| 47 | int init_net(char *file_name) { |
| 48 | FILE *f; |
| 49 | |
| 50 | stringBuffer = malloc(64000); |
| 51 | f = fopen(file_name, "rb"); |
| 52 | if (f == NULL) { |
| 53 | printf("Input file %s not found\n", file_name); |
| 54 | return -1; |
| 55 | } |
| 56 | fscanf(f, "%lld", &words); |
| 57 | fscanf(f, "%lld", &size); |
| 58 | vocab = (char *)malloc((long long)words * max_w * sizeof(char)); |
| 59 | for (a = 0; a < N; a++) bestw[a] = (char *)malloc(max_size * sizeof(char)); |
| 60 | M = (float *)malloc((long long)words * (long long)size * sizeof(float)); |
| 61 | if (M == NULL) { |
| 62 | printf("Cannot allocate memory: %lld MB %lld %lld\n", (long long)words * size * sizeof(float) / 1048576, words, size); |
| 63 | return -1; |
| 64 | } |
| 65 | for (b = 0; b < words; b++) { |
| 66 | a = 0; |
| 67 | while (1) { |
| 68 | vocab[b * max_w + a] = fgetc(f); |
| 69 | if (feof(f) || (vocab[b * max_w + a] == ' ')) break; |
| 70 | if ((a < max_w) && (vocab[b * max_w + a] != '\n')) a++; |
| 71 | } |
| 72 | vocab[b * max_w + a] = 0; |
| 73 | for (a = 0; a < size; a++) fread(&M[a + b * size], sizeof(float), 1, f); |
| 74 | len = 0; |
| 75 | for (a = 0; a < size; a++) len += M[a + b * size] * M[a + b * size]; |
| 76 | len = sqrt(len); |
| 77 | for (a = 0; a < size; a++) M[a + b * size] /= len; |
| 78 | } |
| 79 | fclose(f); |
| 80 | return 0; |
| 81 | } |
| 82 | |
Marc Kupietz | 247500f | 2015-10-09 11:29:01 +0200 | [diff] [blame^] | 83 | SV *print_neighbours(char *st1) { |
Marc Kupietz | dc22b98 | 2015-10-09 09:19:34 +0200 | [diff] [blame] | 84 | FILE *out=stdout; |
| 85 | *stringBuffer=0; |
| 86 | |
| 87 | for (a = 0; a < N; a++) bestd[a] = 0; |
| 88 | for (a = 0; a < N; a++) bestw[a][0] = 0; |
| 89 | a = 0; |
| 90 | cn = 0; |
| 91 | b = 0; |
| 92 | c = 0; |
| 93 | while (1) { |
| 94 | st[cn][b] = st1[c]; |
| 95 | b++; |
| 96 | c++; |
| 97 | st[cn][b] = 0; |
| 98 | if (st1[c] == 0) break; |
| 99 | if (st1[c] == ' ') { |
| 100 | cn++; |
| 101 | b = 0; |
| 102 | c++; |
| 103 | } |
| 104 | } |
| 105 | cn++; |
| 106 | for (a = 0; a < cn; a++) { |
| 107 | for (b = 0; b < words; b++) if (!strcmp(&vocab[b * max_w], st[a])) break; |
| 108 | if (b == words) b = -1; |
| 109 | bi[a] = b; |
| 110 | sprintf(stringBuffer, "\n<pre>Word: \"%s\" Position in vocabulary: %lld</pre>\n", st[a], bi[a]); |
| 111 | if (b == -1) { |
| 112 | sprintf(stringBuffer+strlen(stringBuffer), "Out of dictionary word!\n"); |
| 113 | break; |
| 114 | } |
| 115 | } |
Marc Kupietz | 247500f | 2015-10-09 11:29:01 +0200 | [diff] [blame^] | 116 | if (b == -1) goto end; |
Marc Kupietz | dc22b98 | 2015-10-09 09:19:34 +0200 | [diff] [blame] | 117 | for (a = 0; a < size; a++) vec[a] = 0; |
| 118 | for (b = 0; b < cn; b++) { |
| 119 | if (bi[b] == -1) continue; |
| 120 | for (a = 0; a < size; a++) vec[a] += M[a + bi[b] * size]; |
| 121 | } |
| 122 | len = 0; |
| 123 | for (a = 0; a < size; a++) len += vec[a] * vec[a]; |
| 124 | len = sqrt(len); |
| 125 | for (a = 0; a < size; a++) vec[a] /= len; |
| 126 | for (a = 0; a < N; a++) bestd[a] = -1; |
| 127 | for (a = 0; a < N; a++) bestw[a][0] = 0; |
| 128 | for (c = 0; c < words; c++) { |
| 129 | a = 0; |
| 130 | for (b = 0; b < cn; b++) if (bi[b] == c) a = 1; |
| 131 | if (a == 1) continue; |
| 132 | dist = 0; |
| 133 | for (a = 0; a < size; a++) dist += vec[a] * M[a + c * size]; |
| 134 | for (a = 0; a < N; a++) { |
| 135 | if (dist > bestd[a]) { |
| 136 | for (d = N - 1; d > a; d--) { |
| 137 | bestd[d] = bestd[d - 1]; |
| 138 | strcpy(bestw[d], bestw[d - 1]); |
| 139 | } |
| 140 | bestd[a] = dist; |
| 141 | strcpy(bestw[a], &vocab[c * max_w]); |
| 142 | break; |
| 143 | } |
| 144 | } |
| 145 | } |
Marc Kupietz | 247500f | 2015-10-09 11:29:01 +0200 | [diff] [blame^] | 146 | AV* array = newAV(); |
| 147 | for (a = 0; a < N; a++) { |
| 148 | HV* hash = newHV(); |
| 149 | hv_store(hash, "word", strlen("word"), newSVpvf(bestw[a], 0), 0); |
| 150 | hv_store(hash, "dist", strlen("dist"), newSVnv(bestd[a]), 0); |
| 151 | av_push(array, newRV_noinc((SV*)hash)); |
| 152 | } |
| 153 | end: |
| 154 | return newRV_noinc((SV*)array); |
Marc Kupietz | dc22b98 | 2015-10-09 09:19:34 +0200 | [diff] [blame] | 155 | } |
| 156 | |
| 157 | __DATA__ |
| 158 | |
| 159 | @@ index.html.ep |
| 160 | <!DOCTYPE html> |
| 161 | <html> |
Marc Kupietz | 247500f | 2015-10-09 11:29:01 +0200 | [diff] [blame^] | 162 | <head><title>DeReKo-Word-Vector-Distances</title></head> |
Marc Kupietz | dc22b98 | 2015-10-09 09:19:34 +0200 | [diff] [blame] | 163 | <body> |
Marc Kupietz | 247500f | 2015-10-09 11:29:01 +0200 | [diff] [blame^] | 164 | <p>Word vector model based on a 1.9 billion word sample of DeReKo-2015-II (mainly wikipedia including discussions, current newspapapers and fiction). Trained with <a href="https://code.google.com/p/word2vec/">word2vec</a> using the following parameters:</p> |
| 165 | <pre> |
| 166 | -cbow 1 -size 200 -window 8 -negative 25 -hs 0 -sample 1e-4 -threads 40 -binary 1 -iter 15 |
| 167 | </pre> |
| 168 | </p> |
Marc Kupietz | dc22b98 | 2015-10-09 09:19:34 +0200 | [diff] [blame] | 169 | <form action="<%=url_for('/')->to_abs%>" method="GET"> |
| 170 | Word: <input type="text" name="word"> |
| 171 | <input type="submit" value="Show neighbours"> |
| 172 | </form> |
| 173 | <br> |
Marc Kupietz | 247500f | 2015-10-09 11:29:01 +0200 | [diff] [blame^] | 174 | % if($list) { |
| 175 | <p>Target word: <b><%= $word%></b></p> |
| 176 | <h3>Nearest neighbours</h3> |
| 177 | <table> |
| 178 | <tr> |
| 179 | <th align="right">Pos.</th><th align="left">Word</th><th align="right">Cosine dist.</th> |
| 180 | </tr> |
| 181 | % my $i=1; for my $item (@$list) { |
| 182 | <tr> |
| 183 | <td align="right"> |
| 184 | <%= $i++ %>. |
| 185 | </td> |
| 186 | <td> |
| 187 | <a href="/?word=<%= $item->{word} %>"> |
| 188 | <%= $item->{word} %> |
| 189 | </a> |
| 190 | </td> |
| 191 | <td align="right"> |
| 192 | <%= sprintf("%.3f", $item->{dist}) %> |
| 193 | </td> |
| 194 | </tr> |
| 195 | % } |
| 196 | </table> |
| 197 | % } |
Marc Kupietz | dc22b98 | 2015-10-09 09:19:34 +0200 | [diff] [blame] | 198 | </body> |
| 199 | </html> |
| 200 | |