word2vec

本文主要是介绍word2vec，对大家解决编程问题具有一定的参考价值，需要的程序猿们随着小编来一起学习吧！

在自然语言处理中常常使用预训练的word2vec,来自GoogleNews-vectors-negative300.bin， 下面函数将一句话中的单词转换成词向量，词向量的维度是(300,1), 没有在该word2vec中的单词采用其他的方式，如采用均匀分布，高斯分布等随机初始化

# -*- coding= utf-8 -*-
import numpy as np
# loads 300x1 word vectors from file.
def load_bin_vec(fname, vocab):
    word_vecs = {}
    with open(fname, "rb") as f:
        header = f.readline()
        vocab_size, layer1_size = map(int, header.split()) # 3000000 300
        binary_len = np.dtype('float32').itemsize * layer1_size # 1200
        for line in xrange(vocab_size):
            word = []
            while True:
                ch = f.read(1)
                if ch == ' ':
                    word = ''.join(word)
                    break
                if ch != '\n':
                    word.append(ch)
            if word in vocab:
                word_vecs[word] = np.fromstring(f.read(binary_len), dtype='float32')
            else:
                f.read(binary_len)
    return word_vecs

# add random vectors of unknown words which are not in pre-trained vector file.
# if pre-trained vectors are not used, then initialize all words in vocab with random value.
def add_unknown_words(word_vecs, vocab, min_df=1, k=300):
    for word in vocab:
        if word not in word_vecs and vocab[word] >= min_df:
            word_vecs[word] = np.random.uniform(-0.25, 0.25, k)

vectors_file =  './GoogleNews-vectors-negative300.bin'
vocab=['I', 'can', 'do']

vectors = load_bin_vec(vectors_file, vocab)  # pre-trained vectors
add_unknown_words(vectors, vocab)
print vectors['I']
print '*'*40
print vectors['can']
print '*'*40
print vectors['do']

这篇关于word2vec的文章就介绍到这儿，希望我们推荐的文章对大家有所帮助，也希望大家多多支持为之网！