NNLM原理及Pytorch实现

NNLM

NNLM:Neural Network Language Model,神经网络语言模型。源自Bengio等人于2001年发表在NIPS上的《A Neural Probabilistic Language Model一文。

理论

模型结构

任务

根据\(w_{t-n+1}...w_{t-1}\)来预测\(w_t\)是什么单词,即用\(n-1\)个单词来预测第\(n\)个单词

符号

  • \(V\):词汇的总数,即词汇表的大小
  • \(m\):词向量的长度
  • \(C\)\(V\)行,m列的矩阵表示词向量词表
  • \(C(w)\):单词w的词向量
  • \(d\):隐藏层的偏置
  • \(H\):隐藏层的权重
  • \(U\):隐藏层到输出层的权重
  • \(b\):输出层的偏置
  • \(W\):输入层到输出层的权重
  • \(h\):隐藏层的神经元个数

Data Flow

  1. 获取\(n-1\)个词的词向量,每个词向量的长度是\(m\)
  2. 进行这\(n-1\)个词向量的拼接,形成一个\((n-1)*m\)长度的向量,记做\(X\)
  3. \(X\)送入隐藏层,计算\(hidden_{out}=tanh(X*H+d)\)的到隐藏层的输出
  4. 将隐藏层的输出和输入的词向量同时送入输出层,计算\(y=X*W+hidden_{out}*U+b\),得到输出层\(|V|\)个节点的输出,第\(i\)个节点的输出代表下一个单词是第\(i\)个单词的概率。概率最大的单词为预测到的单词。

代码

Import依赖模块

import torch
import torch.nn as nn
import torch.optim as optim
import torch.utils.data as Data
from torch.autograd import Variable
dtype = torch.FloatTensor

声明变量

sentences = ["i like dog", "i love coffee", "i hate milk"]  # 句子数据集
n_steps = 2  # 用前几个单词来预测下一个单词,e.g. 2个
n_hidden = 2  # 隐藏层的节点个数,e.g. 2个
m = 2  # 词向量的长度

生成词表

word_list = " ".join(sentences).split(" ")  # 获取所有的单词
print("未去重词表:", word_list)
word_list = list(set(word_list))  # 去重
print("去重词表:", word_list)
word_dict = {w: i for i, w in enumerate(word_list)}  # 单词->索引
print("单词索引:", word_dict)
number_dict = {i: w for i, w in enumerate(word_list)}  # 索引->单词
print("索引单词:", number_dict)
num_words = len(word_dict)  # 单词总数
print("单词总数:", num_words)

输出

未去重词表: ['i', 'like', 'dog', 'i', 'love', 'coffee', 'i', 'hate', 'milk']
去重词表: ['coffee', 'love', 'dog', 'like', 'milk', 'hate', 'i']
单词索引: {'coffee': 0, 'love': 1, 'dog': 2, 'like': 3, 'milk': 4, 'hate': 5, 'i': 6}
索引单词: {0: 'coffee', 1: 'love', 2: 'dog', 3: 'like', 4: 'milk', 5: 'hate', 6: 'i'}
单词总数: 7

模型结构

class NNLM(nn.Module):
  # NNLM model architecture
  def __init__(self):
    super(NNLM, self).__init__()
    self.C = nn.Embedding(num_embeddings = num_words, embedding_dim = m)  # 词表
    self.d = nn.Parameter(torch.randn(n_hidden).type(dtype))  # 隐藏层的偏置
    self.H = nn.Parameter(torch.randn(n_steps * m, n_hidden).type(dtype))  # 输入层到隐藏层的权重
    self.U = nn.Parameter(torch.randn(n_hidden, num_words).type(dtype))  # 隐藏层到输出层的权重
    self.b = nn.Parameter(torch.randn(num_words).type(dtype))  # 输出层的偏置
    self.W = nn.Parameter(torch.randn(n_steps * m, num_words).type(dtype))  # 输入层到输出层的权重

  def forward(self, input):
    '''
    input: [batchsize, n_steps] 
    x: [batchsize, n_steps*m]
    hidden_layer: [batchsize, n_hidden]
    output: [batchsize, num_words]
    '''
    x = self.C(input)  # 获得一个batch的词向量的词表
    x = x.view(-1, n_steps * m)
    hidden_out = torch.tanh(torch.mm(x, self.H) + self.d)  # 获取隐藏层输出
    output = torch.mm(x, self.W) + torch.mm(hidden_out, self.U) + self.b  # 获得输出层输出
    return output

格式化输入

def make_batch(sentences):
  '''
  input_batch:一组batch中前n_steps个单词的索引
  target_batch:一组batch中每句话待预测单词的索引
  '''
  input_batch = []
  target_batch = []
  for sentence in sentences:
    word = sentence.split()
    input = [word_dict[w] for w in word[:-1]]
    target = word_dict[word[-1]]
    input_batch.append(input)
    target_batch.append(target)
  return input_batch, target_batch

input_batch, target_batch = make_batch(sentences)
input_batch = torch.LongTensor(input_batch)
target_batch = torch.LongTensor(target_batch)
print("input_batch:", input_batch)
print("target_batch:", target_batch)

输出

input_batch: tensor([[6, 3],
        		     [6, 1],
        		     [6, 5]])
target_batch: tensor([2, 0, 4])

训练

model = NNLM()

criterion = nn.CrossEntropyLoss()  # 使用cross entropy作为loss function
optimizer = optim.Adam(model.parameters(), lr = 0.001)  # 使用Adam作为optimizer

for epoch in range(2000):
  # 梯度清零
  optimizer.zero_grad()
  # 计算predication
  output = model(input_batch)
  # 计算loss
  loss = criterion(output, target_batch)
  if (epoch + 1) % 100 == 0:
    print("Epoch:{}".format(epoch+1), "Loss:{:.3f}".format(loss))
  # 反向传播
  loss.backward()
  # 更新权重参数
  optimizer.step()

输出

Epoch:100 Loss:1.945
Epoch:200 Loss:1.367
Epoch:300 Loss:0.937
Epoch:400 Loss:0.675
Epoch:500 Loss:0.537
Epoch:600 Loss:0.435
Epoch:700 Loss:0.335
Epoch:800 Loss:0.234
Epoch:900 Loss:0.147
Epoch:1000 Loss:0.094
Epoch:1100 Loss:0.065
Epoch:1200 Loss:0.047
Epoch:1300 Loss:0.036
Epoch:1400 Loss:0.029
Epoch:1500 Loss:0.023
Epoch:1600 Loss:0.019
Epoch:1700 Loss:0.016
Epoch:1800 Loss:0.014
Epoch:1900 Loss:0.012
Epoch:2000 Loss:0.011

推理

pred = model(input_batch).data.max(1, keepdim=True)[1]  # 找出概率最大的下标
print("Predict:", pred)
print([sentence.split()[:2] for sentence in sentences], "---->", [number_dict[n.item()] for n in pred.squeeze()])

输出

Predict: tensor([[2],
                       [0],
        	       [4]])
[['i', 'like'], ['i', 'love'], ['i', 'hate']] ----> ['dog', 'coffee', 'milk']

可以和我们的数据集做对比预测准确的。

Reference

posted @ 2021-05-03 09:42  JYRoy  阅读(1740)  评论(0编辑  收藏  举报