Word2Vec代码注解-distance
来源:互联网 发布:如何搭建p2p网络 编辑:程序博客网 时间:2024/06/07 09:54
暂且把文章类型设为“原创”了,如果Mikolov看到的话,还望勿怪。毕竟代码都是别人写的,自己只是往上面加了一些注释,所以,如有什么问题,还请及时通知我改为更合适的选项!
好了,话入正题!
其实,distance的主要功能是计算两个词(组)之间的距离(相似度)。程序将该功能表现为用户输入一条字符串,里面可以包含以空格分割的多个词,然后程序从训练出的模型(demo-word.sh训练出的模型文件名为vectors.bin,二进制文件)中与之最为相似的40个词并输出。以下是添加了注释的源代码,为了与原有代码保持对齐,所有注释均放在代码行后,未另起新行:
// 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 <malloc.h>const long long max_size = 2000; // max length of stringsconst long long N = 40; // number of closest words that will be shownconst long long max_w = 50; // max length of vocabulary entriesint 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];//dist: distance; len: norm of vector; bestd: best distance long long words, size, a, b, c, d, cn, bi[100]; char ch; 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]);//copy filename to the variable f = fopen(file_name, "rb");//open the file in BINARY mode if (f == NULL) { printf("Input file not found\n"); return -1; } fscanf(f, "%lld", &words);//word count fscanf(f, "%lld", &size);//dimension count vocab = (char *)malloc((long long)words * max_w * sizeof(char));//memory alloc, store words, max_w chars per word for (a = 0; a < N; a++) bestw[a] = (char *)malloc(max_size * sizeof(char));//memory alloc, store closest vocabs M = (float *)malloc((long long)words * (long long)size * sizeof(float));//memory alloc, store vectors if (M == NULL) { printf("Cannot allocate memory: %lld MB %lld %lld\n", (long long)words * size * sizeof(float) / 1048576, words, size);//1048576 = 1024 * 1024 return -1; } for (b = 0; b < words; b++) { fscanf(f, "%s%c", &vocab[b * max_w], &ch); 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);//denominator, [ZH_CN]FenMu for (a = 0; a < size; a++) M[a + b * size] /= len;//unit vector, [ZH_CN]DanWei XiangLiang } 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) {//multiple words separated by BLANK symbol st[cn][b] = st1[c]; b++; c++; st[cn][b] = 0;//assign '\0' after the words, if new charactor found, this value can be replaced. if (st1[c] == 0) break; if (st1[c] == ' ') { cn++; b = 0; c++; } } cn++; for (a = 0; a < cn; a++) {//check word position in vocab array. for (b = 0; b < words; b++) if (!strcmp(&vocab[b * max_w], st[a])) break;//found it, b store the position if (b == words) b = -1;//not found, b = -1 bi[a] = b;//position array, store all words input this time 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;//if not found, continue; otherwise, search for closest words 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;//if not found, pass it for (a = 0; a < size; a++) vec[a] += M[a + bi[b] * size];//add all vectors of words found this time } len = 0; for (a = 0; a < size; a++) len += vec[a] * vec[a]; len = sqrt(len);//is it needful? Yes, different vector makes different len value for (a = 0; a < size; a++) vec[a] /= len;//unit vector for (a = 0; a < N; a++) bestd[a] = -1;//best distances, init with -1 for (a = 0; a < N; a++) bestw[a][0] = 0;//best words, init with '\0' for (c = 0; c < words; c++) {//traverse all words a = 0; for (b = 0; b < cn; b++) if (bi[b] == c) a = 1; if (a == 1) continue;//pass words in vocab dist = 0; for (a = 0; a < size; a++) dist += vec[a] * M[a + c * size]; for (a = 0; a < N; a++) {//sort dist, Insertion Sort, DESC 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]);//display results } return 0;}
从代码中可以看出,Mikolov使用的方法原理上还是很简单的。将输入的所有(在模型中存在的)词的向量进行加和(行107),然后进行向量的单位化(行112),最后与vocab中所有在输入序列中出现的word对应的(单位)向量取cosine值。此过程中需要注意的是,Mikolov默认跳过了所有在输入序列中出现的word(行118)。尽管在计算过程中曾经得到过所有词与输入串的dist,但是由于N = 40(行21),因此只有前40条记录会被保存并进行最后的输出,有需要的话可以自行设定该值。
另外,由于max_w = 50(行22),因此对于训练出的结果中存在长词(≥50字节)的情况,还需修改该常量的值。
附上测试截图:
0 0
- Word2Vec代码注解-distance
- 【word2vec】distance.c源码
- word2vec核心代码注释
- word2vec c代码使用说明
- word2vec代码注释
- word2vec代码注释
- word2vec代码注释
- word2vec代码解读
- word2vec的代码注释
- word2vec之TrainModelThread程序详细注解
- word2vec代码注释 - @太极儒
- word2vec
- word2vec
- word2vec
- Word2Vec
- word2vec
- Word2Vec
- Word2Vec
- 2014-第八周项目二-时间类的运算符重载
- Apache负载均衡+tomcat
- android color 属性
- mybatis使用注解2
- 卸载ubuntu自带openJDK,更改成自己的JDK版本
- Word2Vec代码注解-distance
- GPS NMEA码格式
- 冬天的秘密
- Oracle事务基本原理 【转自大话RAC】
- ant build.xml教程详解
- 求复杂多边形面积的算法
- 数据生成器。。。
- placeholder在不同浏览器下的表现及兼容方法
- 关于TNS: could not resolve the connect identifier specified的解决办法