撸一遍代码,加深理解!有些还没有搞懂的部分,希望大家不吝赐教~
源码地址:https://github.com/dav/word2vec/blob/master/src/word2vec.c
// 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 //句子最大长度,超过的进行切分
#define MAX_CODE_LENGTH 40 //huffman编码最大长度
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 { //用于记录单词在构造出的Huffman树中的节点信息
long long cn; //词频
int *point; //从根节点到相应单词所在节点的路径
char *word, *code, codelen; //单词,Huffman编码,编码长度
};
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, cbow = 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;
clock_t start;
int hs = 0, negative = 5;
const int table_size = 1e8;
int *table;
void InitUnigramTable() { //用词频初始化单词的能量表,用于负采样
int a, i;
double train_words_pow = 0;
double d1, power = 0.75; //0.75的指数是为了让低词频的单词,有更多机会被抽中
table = (int *)malloc(table_size * sizeof(int));
for (a = 0; a < vocab_size; a++) train_words_pow += pow(vocab[a].cn, power); //计算总能量
i = 0;
d1 = pow(vocab[i].cn, power) / train_words_pow; //能量比例
for (a = 0; a < table_size; a++) {//table中下一个等分位置
table[a] = i; //vocab_size是去重后的单子个数,table_size是去重前总单词个数(这句不应该出现在下一个if块之后吗?)
if (a / (double)table_size > d1) {
i++; //vocab中下一个单词
d1 += pow(vocab[i].cn, power) / train_words_pow; //加上下一个单词的能量比例
}
if (i >= vocab_size) i = vocab_size - 1; //单词index不能大于或等于vocab_size
}
}
// Reads a single word from a file, assuming space + tab + EOL to be word boundaries
void ReadWord(char *word, FILE *fin) { //从文件流中,读取**一个单词**
int a = 0, ch;
while (!feof(fin)) { //遇到文件结束符返回非0值,否则返回0
ch = fgetc(fin); //从文件中读取一个字符
if (ch == 13) continue; //遇到回车符('\r',ASCII码==13)继续
if ((ch == ' ') || (ch == '\t') || (ch == '\n')) { //单词分隔符
if (a > 0) { //中间遇到换行符,直接跳出循环
if (ch == '\n') ungetc(ch, fin);
break;
}
if (ch == '\n') { //起始位置遇到换行符,返回特定的单词("</s>")
strcpy(word, (char *)"</s>");
return;
} else continue;//所以,"New York"会读为"NewYork"?水平制表符也忽略?
}
word[a] = ch; //读取到的字符,放入word字符数组中
a++; //word字符数组下标+1
if (a >= MAX_STRING - 1) a--; // Truncate too long words
}
word[a] = 0; //word字符数组最后一位,置为空字符
}
// Returns hash value of a word
int GetWordHash(char *word) { //构造一个hash表,快速查找word在vocab中的index
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); //先找到hash表的index
while (1) { //vocab_hash[hash]是单词在vocab中的index
if (vocab_hash[hash] == -1) return -1;
if (!strcmp(word, vocab[vocab_hash[hash]].word)) return vocab_hash[hash]; //如果找到,返回单词在vocab中的index
hash = (hash + 1) % vocab_hash_size;//hash冲突的话,开放寻址法查找
}
return -1;
}
// Reads a word and returns its index in the vocabulary
int ReadWordIndex(FILE *fin) {
char word[MAX_STRING];
ReadWord(word, fin); //从文件中读一个单词
if (feof(fin)) return -1;
return SearchVocab(word); //返回单词在词汇表中的index
}
// 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); //放在vocab末尾
vocab[vocab_size].cn = 0; //词频初始为0,为什么不是1?
vocab_size++; //词汇表size+1
// 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; //添加到hash表中
return vocab_size - 1; //返回的是添加的单词,在vocab中的index
}
// Used later for sorting by word counts
int VocabCompare(const void *a, const void *b) { //比较词频
return ((struct vocab_word *)b)->cn - ((struct vocab_word *)a)->cn;
}
// Sorts the vocabulary by frequency using word counts
void SortVocab() {//按词频排序,并忽略掉词频太低的单词
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); //为何不free整个struct?
} 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; //train_words为参与训练的单词总数(不虑重)
}
}
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_size为实际参与训练的词汇表
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); //问题同上?
vocab_size = b;
for (a = 0; a < vocab_hash_size; a++) vocab_hash[a] = -1;
for (a = 0; a < vocab_size; a++) { //重新hash
// 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++; //每调用ReduceVocab一次,最低词频增加1
}
// 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++) {//需要vocab按词频降序排列
// 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) { //读取文件,按读取顺序形成vocab
ReadWord(word, fin);
if (feof(fin)) break; //文件结束,跳出
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) { //如果该词在vocab中还不存在
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() { //保存文件
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); //将单词和词频写入save_vocab_file文件
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); //读取一个单词
if (feof(fin)) break; //文件结束就跳出
a = AddWordToVocab(word); //添加单词到vocab
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); //把指针放置到文件末尾
file_size = ftell(fin); //返回指针在文件中的位置
fclose(fin);
}
void InitNet() {
long long a, b;
unsigned long long next_random = 1;
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) { //采用分层softmax
a = posix_memalign((void **)&syn1, 128, (long long)vocab_size * layer1_size * sizeof(real));
if (syn1 == 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;
}
if (negative>0) { //采用负采样
a = posix_memalign((void **)&syn1neg, 128, (long long)vocab_size * layer1_size * sizeof(real));
if (syn1neg == 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 < 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;
}
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;
real *neu1 = (real *)calloc(layer1_size, sizeof(real));
real *neu1e = (real *)calloc(layer1_size, sizeof(real));
FILE *fi = fopen(train_file, "rb");
fseek(fi, file_size / (long long)num_threads * (long long)id, SEEK_SET);//文件切分,每个线程负责一部分
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)) { //debug模式,输出更多信息
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)); //学习率随训练单词数的增加,线性减小
if (alpha < starting_alpha * 0.0001) alpha = starting_alpha * 0.0001; //学习率设置下限0.000025
}
if (sentence_length == 0) { //读取一定的单词组合成一个句子
while (1) {
word = ReadWordIndex(fi);
if (feof(fi)) break; //文件流读取结束,跳出
if (word == -1) continue; //这跟上一个if判断条件,有何区别?
word_count++;
if (word == 0) break; //遇到null字符,跳出
// 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; //触发if,即没有被抽到
}
sen[sentence_length] = word; //把单词放进sentence中
sentence_length++;
if (sentence_length >= MAX_SENTENCE_LENGTH) break; //1000个单词组成一个句子?
}
sentence_position = 0; //作为sentence的index,用于对其进行遍历
}
if (feof(fi) || (word_count > train_words / num_threads)) {
word_count_actual += word_count - last_word_count;
local_iter--; //迭代次数-1
if (local_iter == 0) break; //总迭代次数用完,退出
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 < layer1_size; c++) neu1[c] = 0;
for (c = 0; c < layer1_size; c++) neu1e[c] = 0;
next_random = next_random * (unsigned long long)25214903917 + 11;
b = next_random % window;
if (cbow) { //train the cbow architecture
// in -> hidden
cw = 0;
for (a = b; a < window * 2 + 1 - b; a++) if (a != window) {//window是目标单词,2 * (window - b)个单词组成上下文
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]; //layer1_size是单词vec的维度,也是隐含层的神经元数目。neu1是上下文单词平均得到的vec,syn1是对于的参数vec
if (f <= -MAX_EXP) continue;
else if (f >= MAX_EXP) continue;
else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]; //f = sigmoid(neu1 * syn1)
// '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]; //neu1e = neu1e + g * syn1,只累加计算关于neu1的偏导,用于后面的更新
// Learn weights hidden -> output
for (c = 0; c < layer1_size; c++) syn1[c + l2] += g * neu1[c]; //syn1 = syn1 + alpha * [ (1 - vocab[word].code[d] - f) * syn1 ],直接更新syn1
}
// NEGATIVE SAMPLING
if (negative > 0) for (d = 0; d < negative + 1; d++) { //只对(负样本+1)个单词进行更新
if (d == 0) {
target = word;
label = 1; //标记为正样本
} else {
next_random = next_random * (unsigned long long)25214903917 + 11;
target = table[(next_random >> 16) % table_size];
if (target == 0) target = next_random % (vocab_size - 1) + 1; //不能采样到词汇表第1个单词,该位置存放的是特殊字符
if (target == word) continue; //也不能采样到正样本
label = 0; //标记为负样本
}
l2 = target * layer1_size; //该单词的参数,对应于参数列表中的syn1neg[target * layer1_size : target * layer1_size + layer1_size - 1]
f = 0;
for (c = 0; c < layer1_size; c++) f += neu1[c] * syn1neg[c + l2]; //上下文中目标单词的vec * 抽样集合中目标单词的参数vec
if (f > MAX_EXP) g = (label - 1) * alpha; //f > MAX_EXP时,exp(f) := 1
else if (f < -MAX_EXP) g = (label - 0) * alpha; //f < -MAX_EXP时,exp(f) := 0
else g = (label - expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]) * alpha; //-MAX_EXP < f < MAX_EXP时,exp(f)查表
for (c = 0; c < layer1_size; c++) neu1e[c] += g * syn1neg[c + l2]; //对抽样集合中所有单词,累加目标函数对该单词vec的偏导,用于对上下文中单词vec的更新
for (c = 0; c < layer1_size; c++) syn1neg[c + l2] += g * neu1[c]; //更新抽样集合中目标单词的参数vec
}
// 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]; //更新上下文中的每个单词的vec
}
}
} else { //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++) { //对路径上的每一个节点
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]; //syn0(c)是目标单词的vec,syn1(c, d)是参数的vec
if (f <= -MAX_EXP) continue;
else if (f >= MAX_EXP) continue;
else f = expTable[(int)((f + MAX_EXP) * (EXP_TABLE_SIZE / MAX_EXP / 2))]; //f=sigmoid(syn0 * syn1)
// '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]; //neu1e = neu1e + g * syn1,只累加计算关于neu1的偏导,用于后面的更新
// Learn weights hidden -> output
for (c = 0; c < layer1_size; c++) syn1[c + l2] += g * syn0[c + l1]; //syn1 = syn1 + alpha * [ (1 - vocab[word].code[d] - f) * syn0 ],直接更新syn1
}
// 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;
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]; //更新上下文中的每个单词的vec
}
}
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(); //初始化网络
if (negative > 0) InitUnigramTable(); //如果负采样,初始化能量表
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);//以二进制方式向文件流中写入数据
else for (b = 0; b < layer1_size; b++) fprintf(fo, "%lf ", syn0[a * layer1_size + b]);//格式化输出到一个流/文件中
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) { //解析外部输入参数
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-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-cbow <int>\n");
printf("\t\tUse the continuous bag of words model; default is 1 (use 0 for skip-gram model)\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 -cbow 1 -iter 3\n\n");
return 0;
}
output_file[0] = 0;
save_vocab_file[0] = 0;
read_vocab_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 *)"-cbow", argc, argv)) > 0) cbow = atoi(argv[i + 1]);
if (cbow) alpha = 0.05; //cbow的初始学习率为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 *)"-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]);
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;
}
参考文献
[1] word2vec原理推导与代码分析
[2] word2vec源码解析之word2vec.c
[3] word2vec Parameter Learning Explained, Xin Rong