pytorch --- word2vec 实现 --《Efficient Estimation of Word Representations in Vector Space》

论文来自Mikolov等人的《Efficient Estimation of Word Representations in Vector Space》

论文地址： 66666

论文介绍了2个方法，原理不解释...

skim code and comment https://github.com/graykode/nlp-tutorial:

# -*- coding: utf-8 -*-
# @time : 2019/11/9  12:53

import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from torch.autograd import Variable
import matplotlib.pyplot as plt

dtype = torch.FloatTensor

# 3 Words Sentence
sentences = [ "i like dog", "i like cat", "i like animal",
              "dog cat animal", "apple cat dog like", "dog fish milk like",
              "dog cat eyes like", "i like apple", "apple i hate",
              "apple i movie book music like", "cat dog hate", "cat dog like"]

word_sequence = " ".join(sentences).split()
word_list = " ".join(sentences).split()
word_list = list(set(word_list))
word_dict = {w: i for i, w in enumerate(word_list)}

# Word2Vec Parameter
batch_size = 20  # To show 2 dim embedding graph
embedding_size = 2  # To show 2 dim embedding graph
voc_size = len(word_list)

# 产生 batch_size个，每个都是一个input和label, both are ont-hot vector
def random_batch(data, size):
    random_inputs = []
    random_labels = []
    random_index = np.random.choice(range(len(data)), size, replace=False)

    for i in random_index:
        random_inputs.append(np.eye(voc_size)[data[i][0]])  # target
        random_labels.append(data[i][1])  # context word

    return random_inputs, random_labels

# Make skip gram of one size window
skip_grams = []
# 从第2个word_sequence开始(index=1),预测index=0和index=2，也就是[index=1,index=0]和[index=1,index=2]的添加到skim_grams中
for i in range(1, len(word_sequence) - 1):
    target = word_dict[word_sequence[i]]
    context = [word_dict[word_sequence[i - 1]], word_dict[word_sequence[i + 1]]]

    for w in context:
        skip_grams.append([target, w])

# Model
class Word2Vec(nn.Module):
    def __init__(self):
        super(Word2Vec, self).__init__()

        # W and WT is not Traspose relationship
        self.W = nn.Parameter(-2 * torch.rand(voc_size, embedding_size) + 1).type(dtype) # voc_size > embedding_size Weight
        self.WT = nn.Parameter(-2 * torch.rand(embedding_size, voc_size) + 1).type(dtype) # embedding_size > voc_size Weight

    def forward(self, X):
        # X : [batch_size, voc_size]
        hidden_layer = torch.matmul(X, self.W) # hidden_layer : [batch_size, embedding_size]
        output_layer = torch.matmul(hidden_layer, self.WT) # output_layer : [batch_size, voc_size]
        return output_layer

model = Word2Vec()

criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.001)

# Training
for epoch in range(5000):

    input_batch, target_batch = random_batch(skip_grams, batch_size)

    input_batch = Variable(torch.Tensor(input_batch))
    target_batch = Variable(torch.LongTensor(target_batch))

    optimizer.zero_grad()
    output = model(input_batch)

    # output : [batch_size, voc_size], target_batch : [batch_size] (LongTensor, not one-hot)
    loss = criterion(output, target_batch)
    if (epoch + 1)%1000 == 0:
        print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.6f}'.format(loss))

    loss.backward()
    optimizer.step()

# because
# input_size is [batch_size,voc_size] , ( a word is one-hot voctor(lenght is voc_size) )
# W is [voc_size,emmedding_size]
# a word*W ,result is same as:
# [1,0,0]*[w1,w4
#          w2,w5
#          w3,w6]
# so one word embedding vector is [w1,w4]
# 即: W[i][0],W[i][1]
for i, label in enumerate(word_list):
    W, WT = model.parameters()
    x,y = float(W[i][0]), float(W[i][1])
    plt.scatter(x, y)
    plt.annotate(label, xy=(x, y), xytext=(5, 2), textcoords='offset points', ha='right', va='bottom')
plt.show()