TextCNN 的 PyTorch 实现

B 站视频讲解

本文主要介绍一篇将 CNN 应用到 NLP 领域的一篇论文 Convolutional Neural Networks for Sentence Classification，然后给出 PyTorch 实现

论文比较短，总体流程也不复杂，最主要的是下面这张图，只要理解了这张图，就知道如何写代码了。如果你不了解 CNN，请先看我的这篇文章 CS231n 笔记：通俗理解 CNN

下图的 feature map 是将一句话中的各个词通过 WordEmbedding 得到的，feature map 的宽为 embedding 的维度，长为一句话的单词数量。例如下图中，很明显就是用一个 6 维的向量去编码每个词，并且一句话中有 9 个词

之所以有两张 feature map，你可以理解为 batchsize 为 2

其中，红色的框代表的就是卷积核。而且很明显可以看出，这是一个长宽不等的卷积核。有意思的是，卷积核的宽可以认为是 n-gram，比方说下图卷积核宽为 2，所以同时考虑了 "wait" 和 "for" 两个单词的词向量，因此可以认为该卷积是一个类似于 bigram 的模型

后面的部分就是传统 CNN 的步骤，激活、池化、Flatten，没什么好说的

代码实现（PyTorch 版）

源码来自于 nlp-tutorial，我在其基础上进行了修改（原本的代码感觉有很多问题）

'''
  code by Tae Hwan Jung(Jeff Jung) @graykode, modify by wmathor
'''
import torch
import numpy as np
import torch.nn as nn
import torch.optim as optim
import torch.utils.data as Data
import torch.nn.functional as F

dtype = torch.FloatTensor
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

下面代码就是定义一些数据，以及设置一些常规参数

# 3 words sentences (=sequence_length is 3)
sentences = ["i love you", "he loves me", "she likes baseball", "i hate you", "sorry for that", "this is awful"]
labels = [1, 1, 1, 0, 0, 0]  # 1 is good, 0 is not good.

# TextCNN Parameter
embedding_size = 2
sequence_length = len(sentences[0]) # every sentences contains sequence_length(=3) words
num_classes = len(set(labels))  # num_classes=2
batch_size = 3

word_list = " ".join(sentences).split()
vocab = list(set(word_list))
word2idx = {w: i for i, w in enumerate(vocab)}
vocab_size = len(vocab)

数据预处理

def make_data(sentences, labels):
  inputs = []
  for sen in sentences:
      inputs.append([word2idx[n] for n in sen.split()])

  targets = []
  for out in labels:
      targets.append(out) # To using Torch Softmax Loss function
  return inputs, targets

input_batch, target_batch = make_data(sentences, labels)
input_batch, target_batch = torch.LongTensor(input_batch), torch.LongTensor(target_batch)

dataset = Data.TensorDataset(input_batch, target_batch)
loader = Data.DataLoader(dataset, batch_size, True)

构建模型

class TextCNN(nn.Module):
    def __init__(self):
        super(TextCNN, self).__init__()
        self.W = nn.Embedding(vocab_size, embedding_size)
        output_channel = 3
        self.conv = nn.Sequential(
            # conv : [input_channel(=1), output_channel, (filter_height, filter_width), stride=1]
            nn.Conv2d(1, output_channel, (2, embedding_size)),
            nn.ReLU(),
            # pool : ((filter_height, filter_width))
            nn.MaxPool2d((2, 1)),
        )
        # fc
        self.fc = nn.Linear(output_channel, num_classes)

    def forward(self, X):
      '''
      X: [batch_size, sequence_length]
      '''
      batch_size = X.shape[0]
      embedding_X = self.W(X) # [batch_size, sequence_length, embedding_size]
      embedding_X = embedding_X.unsqueeze(1) # add channel(=1) [batch, channel(=1), sequence_length, embedding_size]
      conved = self.conv(embedding_X) # [batch_size, output_channel*1*1]
      flatten = conved.view(batch_size, -1)
      output = self.fc(flatten)
      return output

下面详细介绍一下数据在网络中流动的过程中维度的变化。输入数据是个矩阵，矩阵维度为 [batch_size, seqence_length]，输入矩阵的数字代表的是某个词在整个词库中的索引（下标）

首先通过 Embedding 层，也就是查表，将每个索引转为一个向量，比方说 12 可能会变成 [0.3,0.6,0.12,...]，因此整个数据无形中就增加了一个维度，变成了 [batch_size, sequence_length, embedding_size]

之后使用 unsqueeze(1) 函数使数据增加一个维度，变成 [batch_size, 1, sequence_length, embedding_size]。现在的数据才能做卷积，因为在传统 CNN 中，输入数据就应该是 [batch_size, in_channel, height, width] 这种维度

[batch_size, 1, 3, 2] 的输入数据通过 nn.Conv2d(1, 3, (2, 2)) 的卷积之后，得到的就是 [batch_size, 3, 2, 1] 的数据，由于经过 ReLU 激活函数是不改变维度的，所以就没画出来。最后经过一个 nn.MaxPool2d((2, 1)) 池化，得到的数据维度就是 [batch_size, 3, 1, 1]

训练

model = TextCNN().to(device)
criterion = nn.CrossEntropyLoss().to(device)
optimizer = optim.Adam(model.parameters(), lr=1e-3)

# Training
for epoch in range(5000):
  for batch_x, batch_y in loader:
    batch_x, batch_y = batch_x.to(device), batch_y.to(device)
    pred = model(batch_x)
    loss = criterion(pred, batch_y)
    if (epoch + 1) % 1000 == 0:
        print('Epoch:', '%04d' % (epoch + 1), 'loss =', '{:.6f}'.format(loss))

    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

测试

# Test
test_text = 'i hate me'
tests = [[word2idx[n] for n in test_text.split()]]
test_batch = torch.LongTensor(tests).to(device)
# Predict
model = model.eval()
predict = model(test_batch).data.max(1, keepdim=True)[1]
if predict[0][0] == 0:
    print(test_text,"is Bad Mean...")
else:
    print(test_text,"is Good Mean!!")

完整代码如下：

'''
  code by Tae Hwan Jung(Jeff Jung) @graykode, modify by wmathor
'''
import torch
import numpy as np
import torch.nn as nn
import torch.optim as optim
import torch.utils.data as Data
import torch.nn.functional as F

dtype = torch.FloatTensor
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# 3 words sentences (=sequence_length is 3)
sentences = ["i love you", "he loves me", "she likes baseball", "i hate you", "sorry for that", "this is awful"]
labels = [1, 1, 1, 0, 0, 0]  # 1 is good, 0 is not good.

# TextCNN Parameter
embedding_size = 2
sequence_length = len(sentences[0]) # every sentences contains sequence_length(=3) words
num_classes = 2  # 0 or 1
batch_size = 3

word_list = " ".join(sentences).split()
vocab = list(set(word_list))
word2idx = {w: i for i, w in enumerate(vocab)}
vocab_size = len(vocab)

def make_data(sentences, labels):
  inputs = []
  for sen in sentences:
      inputs.append([word2idx[n] for n in sen.split()])

  targets = []
  for out in labels:
      targets.append(out) # To using Torch Softmax Loss function
  return inputs, targets

input_batch, target_batch = make_data(sentences, labels)
input_batch, target_batch = torch.LongTensor(input_batch), torch.LongTensor(target_batch)

dataset = Data.TensorDataset(input_batch, target_batch)
loader = Data.DataLoader(dataset, batch_size, True)

class TextCNN(nn.Module):
    def __init__(self):
        super(TextCNN, self).__init__()
        self.W = nn.Embedding(vocab_size, embedding_size)
        output_channel = 3
        self.conv = nn.Sequential(
            # conv : [input_channel(=1), output_channel, (filter_height, filter_width), stride=1]
            nn.Conv2d(1, output_channel, (2, embedding_size)),
            nn.ReLU(),
            # pool : ((filter_height, filter_width))
            nn.MaxPool2d((2, 1)),
        )
        # fc
        self.fc = nn.Linear(output_channel, num_classes)

    def forward(self, X):
      '''
      X: [batch_size, sequence_length]
      '''
      batch_size = X.shape[0]
      embedding_X = self.W(X) # [batch_size, sequence_length, embedding_size]
      embedding_X = embedding_X.unsqueeze(1) # add channel(=1) [batch, channel(=1), sequence_length, embedding_size]
      conved = self.conv(embedding_X) # [batch_size, output_channel, 1, 1]
      flatten = conved.view(batch_size, -1) # [batch_size, output_channel*1*1]
      output = self.fc(flatten)
      return output

model = TextCNN().to(device)
criterion = nn.CrossEntropyLoss().to(device)
optimizer = optim.Adam(model.parameters(), lr=1e-3)

# Training
for epoch in range(5000):
  for batch_x, batch_y in loader:
    batch_x, batch_y = batch_x.to(device), batch_y.to(device)
    pred = model(batch_x)
    loss = criterion(pred, batch_y)
    if (epoch + 1) % 1000 == 0:
        print('Epoch:', '%04d' % (epoch + 1), 'loss =', '{:.6f}'.format(loss))

    optimizer.zero_grad()
    loss.backward()
    optimizer.step()
    
# Test
test_text = 'i hate me'
tests = [[word2idx[n] for n in test_text.split()]]
test_batch = torch.LongTensor(tests).to(device)
# Predict
model = model.eval()
predict = model(test_batch).data.max(1, keepdim=True)[1]
if predict[0][0] == 0:
    print(test_text,"is Bad Mean...")
else:
    print(test_text,"is Good Mean!!")