我可以从word2vec网站下载GoogleNews-vectors-negative300.bin.gz。 .bin文件(大约3.4GB)是一种对我没用的二进制格式。 Tomas Mikolov assures us"将二进制格式转换为文本格式应该相当简单(尽管这需要更多的磁盘空间)。检查距离工具中的代码,读取二进制文件相当简单。"不幸的是,我不太了解C http://word2vec.googlecode.com/svn/trunk/distance.c。
据说gensim也可以这样做,但我发现的所有教程似乎都是关于从文本转换,而不是其他方式。
有人可以建议修改C代码或gensim发出文本的说明吗?
答案 0 :(得分:75)
我使用此代码加载二进制模型,然后将模型保存到文本文件
from gensim.models.keyedvectors import KeyedVectors
model = KeyedVectors.load_word2vec_format('path/to/GoogleNews-vectors-negative300.bin', binary=True)
model.save_word2vec_format('path/to/GoogleNews-vectors-negative300.txt', binary=False)
注意:
以上代码适用于gensim的新版本。对于之前的版本,我使用了此代码:
from gensim.models import word2vec
model = word2vec.Word2Vec.load_word2vec_format('path/to/GoogleNews-vectors-negative300.bin', binary=True)
model.save_word2vec_format('path/to/GoogleNews-vectors-negative300.txt', binary=False)
答案 1 :(得分:17)
在word2vec-toolkit邮件列表上,Thomas Mensink以小型C程序的形式提供了answer,它将.bin文件转换为文本。这是distance.c文件的修改。我用下面的Thomas代码替换了原来的distance.c并重建了word2vec(make clean; make),并将编译后的距离重命名为readbin。然后./readbin vector.bin
将创建vector.bin的文本版本。
// 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 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 file_name[max_size];
float len;
long long words, size, a, b;
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]);
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++) {
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);
for (a = 0; a < size; a++) M[a + b * size] /= len;
}
fclose(f);
//Code added by Thomas Mensink
//output the vectors of the binary format in text
printf("%lld %lld #File: %s\n",words,size,file_name);
for (a = 0; a < words; a++){
printf("%s ",&vocab[a * max_w]);
for (b = 0; b< size; b++){ printf("%f ",M[a*size + b]); }
printf("\b\b\n");
}
return 0;
}
我从printf
删除了“\ b \ b”。
顺便说一下,生成的文本文件仍然包含文本词和一些不必要的空格,我不想进行一些数值计算。我使用bash命令从每行删除了初始文本列和尾随空白。
cut --complement -d ' ' -f 1 GoogleNews-vectors-negative300.txt > GoogleNews-vectors-negative300_tuples-only.txt
sed 's/ $//' GoogleNews-vectors-negative300_tuples-only.txt
答案 2 :(得分:7)
格式为IEEE 754单精度二进制浮点格式:binary32 http://en.wikipedia.org/wiki/Single-precision_floating-point_format 他们使用little-endian。
举个例子:
下一行包括词汇 首先是字,然后是(300 * 4字节的浮点值,每个4字节) 尺寸):
getByte till byte==32 (space). (60 47 115 62 32 => <\s>[space])
然后每个接下来的4个字节将代表一个浮点数
下一个4字节:0 0 -108 58 =&gt; 0.001129150390625。
您可以查看维基百科链接以了解具体方法,让我以此为例:
(little-endian - &gt;逆序)00111010 10010100 00000000 00000000
value = sign * exp * pre
答案 3 :(得分:5)
您可以在word2vec中加载二进制文件,然后保存文本版本,如下所示:
from gensim.models import word2vec
model = word2vec.Word2Vec.load_word2vec_format('Path/to/GoogleNews-vectors-negative300.bin', binary=True)
model.save("file.txt")
`
答案 4 :(得分:4)
我正在使用gensim处理GoogleNews-vectors-negative300.bin,并在加载模型时加入binary = True
标记。
from gensim import word2vec
model = word2vec.Word2Vec.load_word2vec_format('Path/to/GoogleNews-vectors-negative300.bin', binary=True)
似乎工作正常。
答案 5 :(得分:2)
我有类似的问题,我希望将bin / non-bin(gensim)模型输出为CSV。
这是在python上执行该操作的代码,它假设您已安装gensim:
答案 6 :(得分:2)
convertvec是一个小工具,可以为word2vec库转换不同格式的矢量。
将矢量从二进制转换为纯文本:
./ convertvec bin2txt input.bin output.txt
将矢量从纯文本转换为二进制文件:
./ convertvec txt2bin input.txt output.bin
答案 7 :(得分:2)
如果您收到错误:
ImportError: No module named models.word2vec
那是因为有一个API更新。这将有效:
from gensim.models.keyedvectors import KeyedVectors
model = KeyedVectors.load_word2vec_format('./GoogleNews-vectors-negative300.bin', binary=True)
model.save_word2vec_format('./GoogleNews-vectors-negative300.txt', binary=False)
答案 8 :(得分:1)
以下是我使用的代码:
import codecs
from gensim.models import Word2Vec
def main():
path_to_model = 'GoogleNews-vectors-negative300.bin'
output_file = 'GoogleNews-vectors-negative300_test.txt'
export_to_file(path_to_model, output_file)
def export_to_file(path_to_model, output_file):
output = codecs.open(output_file, 'w' , 'utf-8')
model = Word2Vec.load_word2vec_format(path_to_model, binary=True)
print('done loading Word2Vec')
vocab = model.vocab
for mid in vocab:
#print(model[mid])
#print(mid)
vector = list()
for dimension in model[mid]:
vector.append(str(dimension))
#line = { "mid": mid, "vector": vector }
vector_str = ",".join(vector)
line = mid + "\t" + vector_str
#line = json.dumps(line)
output.write(line + "\n")
output.close()
if __name__ == "__main__":
main()
#cProfile.run('main()') # if you want to do some profiling
答案 9 :(得分:0)
快速更新,因为现在有更简单的方法。
如果您使用https://github.com/dav/word2vec中的word2vec
,则会有一个名为-binary
的附加选项接受1
生成二进制文件或0
生成文本文件。此示例来自repo中的demo-word.sh
:
time ./word2vec -train text8 -output vectors.bin -cbow 1 -size 200 -window 8 -negative 25 -hs 0 -sample 1e-4 -threads 20 -binary 0 -iter 15