Word2vec的高速化：CBOW模型 24.4.17打卡

word2vec的高速化

1.CBOW模型的学习代码

import sys
sys.path.append('..')
from common import config
# 在用GPU运行时，请打开下面的注释（需要cupy）
# ===============================================
#config.GPU = True
# ===============================================
from common.np import *
import pickle
from common.trainer import Trainer
from common.optimizer import Adam
from cbow import CBOW
from skip_gram import SkipGram
from common.util import create_contexts_target, to_cpu, to_gpu
from dataset import ptb


# 设定超参数
window_size = 5
hidden_size = 100
batch_size = 100
max_epoch = 10

# 读入数据
corpus, word_to_id, id_to_word = ptb.load_data('train')		#加载熟练数据
vocab_size = len(word_to_id)

contexts, target = create_contexts_target(corpus, window_size)		#
if config.GPU:
    contexts, target = to_gpu(contexts), to_gpu(target)

# 生成模型等
model = CBOW(vocab_size, hidden_size, window_size, corpus)
# model = SkipGram(vocab_size, hidden_size, window_size, corpus)
optimizer = Adam()
trainer = Trainer(model, optimizer)

# 开始学习
trainer.fit(contexts, target, max_epoch, batch_size)
trainer.plot()

# 保存必要数据，以便后续使用
word_vecs = model.word_vecs
if config.GPU:
    word_vecs = to_cpu(word_vecs)
params = {}
params['word_vecs'] = word_vecs.astype(np.float16)
params['word_to_id'] = word_to_id
params['id_to_word'] = id_to_word
pkl_file = 'cbow_params.pkl'  # or 'skipgram_params.pkl'
with open(pkl_file, 'wb') as f:
    pickle.dump(params, f, -1)

2.create_contexts_target(生成上下文和目标词的函数)

def create_contexts_target(corpus, window_size=1):
    '''生成上下文和目标词

    :param corpus: 语料库（单词ID列表）
    :param window_size: 窗口大小（当窗口大小为1时，左右各1个单词为上下文）
    :return:
    '''
    target = corpus[window_size:-window_size]
    contexts = []

    for idx in range(window_size, len(corpus)-window_size):
        cs = []
        for t in range(-window_size, window_size + 1):
            if t == 0:
                continue
            cs.append(corpus[idx + t])
        contexts.append(cs)

    return np.array(contexts), np.array(target)

CBOW模型的实现代码：

import sys
sys.path.append('..')
from common.np import *  # import numpy as np
from common.layers import Embedding
from ch04.negative_sampling_layer import NegativeSamplingLoss


class CBOW:
    def __init__(self, vocab_size, hidden_size, window_size, corpus):
        V, H = vocab_size, hidden_size

        # 初始化权重
        W_in = 0.01 * np.random.randn(V, H).astype('f')
        W_out = 0.01 * np.random.randn(V, H).astype('f')

        # 生成层
        self.in_layers = []
        for i in range(2 * window_size):		#创建 2 * window_size 个Embedding 层
            layer = Embedding(W_in)  # 使用Embedding层，输入层-隐藏层
            self.in_layers.append(layer)
        self.ns_loss = NegativeSamplingLoss(W_out, corpus, power=0.75, sample_size=5)

        # 将所有的权重和梯度整理到列表中
        layers = self.in_layers + [self.ns_loss]
        self.params, self.grads = [], []
        for layer in layers:
            self.params += layer.params
            self.grads += layer.grads

        # 将单词的分布式表示设置为成员变量
        self.word_vecs = W_in
	#contexts, target如图所示
    def forward(self, contexts, target):
        h = 0
        for i, layer in enumerate(self.in_layers):
            h += layer.forward(contexts[:, i])
        h *= 1 / len(self.in_layers)
        loss = self.ns_loss.forward(h, target)		#实现负采样，NegativeSamplingLoss，并返回损失函数
        return loss

    def backward(self, dout=1):
        dout = self.ns_loss.backward(dout)
        dout *= 1 / len(self.in_layers)
        for layer in self.in_layers:
            layer.backward(dout)
        return None

forward：contexts(文本)是目标词的上下文的关联词，target（目标）是我们需要预测的值

这里我们先给出正确解题思路：（思维框架）

我们首先要调用输入层的Embedding，相关词经过Embedding（也就是降低数据处理量，简化），然后，再相加，再除以相关词的数量，得到平均数，这就求出了隐藏层

然后再调用损失函数，即可求出损失值啦。