word2vec bin 파일을 텍스트로 변환

word2vec bin 파일을 텍스트로 변환

---

word2vec 사이트 에서 GoogleNews-vectors-negative300.bin.gz를 다운로드 할 수 있습니다. .bin 파일 (약 3.4GB)은 나에게 유용하지 않은 바이너리 형식입니다. 토마스 Mikolov은 우리를 보장 "(즉, 더 많은 디스크 공간을 차지합니다하지만) 텍스트 형식으로 바이너리 형식을 변환하는 매우 간단합니다. 거리 도구의 코드를 확인,이 바이너리 파일을 읽을 오히려 사소한."고 불행히도 나는 http://word2vec.googlecode.com/svn/trunk/distance.c 를 이해하기에 충분한 C를 모릅니다 .

기발한 gensim 또한이 작업을 수행하지만, 내가 찾은 모든 튜토리얼 변환하는 방법에 대한 것으로 보일 수 에서 , 텍스트가 아닌 다른 방법을.

누군가가 gensim이 텍스트를 내보내도록 C 코드를 수정하거나 지침을 제안 할 수 있습니까?

---

이 코드를 사용하여 바이너리 모델을로드 한 다음 모델을 텍스트 파일에 저장합니다.

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)

참조 : API 및 nullege .

노트 :

위 코드는 gensim의 새 버전입니다. 들어 이전 버전,이 코드를 사용 :

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)

---

word2vec-toolkit 메일 링리스트에서 Thomas Mensink는 .bin 파일을 텍스트로 변환하는 작은 C 프로그램의 형태로 답변 을 제공했습니다 . distance.c 파일을 수정 한 것입니다. 원래 distance.c를 아래의 Thomas의 코드로 바꾸고 word2vec (make clean; make)를 다시 작성하고 컴파일 된 distance의 이름을 readbin으로 변경했습니다. 그런 다음 ./readbin vector.binvector.bin의 텍스트 버전을 만듭니다.

//  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;
}

I removed the "\b\b" from the printf.

By the way, the resulting text file still contained the text word and some unnecessary whitespace which I did not want for some numerical calculations. I removed the initial text column and the trailing blank from each line with bash commands.

cut --complement -d ' ' -f 1 GoogleNews-vectors-negative300.txt > GoogleNews-vectors-negative300_tuples-only.txt
sed 's/ $//' GoogleNews-vectors-negative300_tuples-only.txt

---

the format is IEEE 754 single-precision binary floating-point format: binary32 http://en.wikipedia.org/wiki/Single-precision_floating-point_format They use little-endian.

Let do an example:

• First line is string format: "3000000 300\n" (vocabSize & vecSize, getByte till byte=='\n')

• Next line include the vocab word first, and then (300*4 byte of float value, 4 byte for each dimension):

  getByte till byte==32 (space). (60 47 115 62 32 => <\s>[space])
  

• then each next 4 byte will represent one float number

  next 4 byte: 0 0 -108 58 => 0.001129150390625.

You can check the wikipedia link to see how, let me do this one as example:

(little-endian -> reverse order) 00111010 10010100 00000000 00000000

• first is sign bit => sign = 1 (else = -1)
• next 8 bits => 117 => exp = 2^(117-127)
• next 23 bits => pre = 0*2^(-1) + 0*2^(-2) + 1*2^(-3) + 1*2^(-5)

value = sign * exp * pre

---

You can load the binary file in word2vec, and then save the text version like this:

from gensim.models import word2vec
 model = word2vec.Word2Vec.load_word2vec_format('Path/to/GoogleNews-vectors-negative300.bin', binary=True)
 model.save("file.txt")

`

---

I am using gensim to work with the GoogleNews-vectors-negative300.bin and I am including a binary = True flag while loading the model.

from gensim import word2vec

model = word2vec.Word2Vec.load_word2vec_format('Path/to/GoogleNews-vectors-negative300.bin', binary=True) 

Seems to be working fine.

---

I had a similar issue, I wanted to get bin/non-bin(gensim) models output as CSV.

here is the code which does that on python, it assumes you have gensim installed:

https://gist.github.com/dav009/10a742de43246210f3ba

---

Here is the code I use:

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

---

convertvec is a small tool to convert vectors between different formats for the word2vec library.

Convert vectors from binary to plain text:

./convertvec bin2txt input.bin output.txt

Convert vectors from plain text to binary:

./convertvec txt2bin input.txt output.bin

---

If you get the Error:

ImportError: No module named models.word2vec

then it is because there was an API update. This will work:

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)

---

Just a quick update as now there is easier way.

If you are using word2vec from https://github.com/dav/word2vec there is additional option called -binary which accept 1 to generate binary file or 0 to generate text file. This example comes from demo-word.sh in the repo:

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

참고URL : https://stackoverflow.com/questions/27324292/convert-word2vec-bin-file-to-text