PyTorch实战：基于Seq2seq模型处理机器翻译任务(模型预测)
文章目录
- 引言
- 数据预处理
-
- 加载字典对象`en2id`和`zh2id`
- 文本分词
- 加载训练好的Seq2Seq模型
- 模型预测完整代码
- 结束语
引言
随着全球化的深入，翻译需求日益增长。传统的人工翻译方式虽然质量高，但效率低，成本高。机器翻译的出现，为解决这一问题提供了可能。英译中机器翻译任务是机器翻译领域的一个重要分支，旨在将英文文本自动翻译成中文。本博客以《PyTorch自然语言处理入门与实战》第九章的Seq2seq模型处理英译中翻译任务作为基础，附上模型预测模块。
模型的训练及验证模块的详细解析见PyTorch实战：基于Seq2seq模型处理机器翻译任务(模型训练及验证)
数据预处理
加载字典对象en2id和zh2id
在预测阶段中，需要加载模型训练及验证阶段保存的字典对象en2id和zh2id。
代码如下：
import pickle
with open("en2id.pkl", 'rb') as f:
en2id = pickle.load(f)
with open("zh2id.pkl", 'rb') as f:
zh2id = pickle.load(f)
文本分词
在对输入文本进行预测时，需要先将文本进行分词操作。参考代码如下：
def extract_words(sentence):
"""
从给定的英文句子中提取单词，并去除单词后的标点符号。
Args:
sentence (str): 要提取单词的英文句子。
Returns:
List[str]: 提取并处理后的单词列表。
"""
en_words = []
for w in sentence.split(' '): # 将英文句子按空格分词
w = w.replace('.', '').replace(',', '') # 去除跟单词连着的标点符号
w = w.lower() # 统一单词大小写
if w:
en_words.append(w)
return en_words
# 测试函数
sentence = 'I am Dave Gallo.'
print(extract_words(sentence))
运行结果：
加载训练好的Seq2Seq模型
代码如下：
import torch
import torch.nn as nn
class Encoder(nn.Module):
def __init__(self, input_dim, emb_dim, hid_dim, n_layers, dropout):
super().__init__()
self.hid_dim = hid_dim
self.n_layers = n_layers
self.embedding = nn.Embedding(input_dim, emb_dim) # 词嵌入
self.rnn = nn.LSTM(emb_dim, hid_dim, n_layers, dropout=dropout)
self.dropout = nn.Dropout(dropout)
def forward(self, src):
# src = (src len, batch size)
embedded = self.dropout(self.embedding(src))
# embedded = (src len, batch size, emb dim)
outputs, (hidden, cell) = self.rnn(embedded)
# outputs = (src len, batch size, hid dim * n directions)
# hidden = (n layers * n directions, batch size, hid dim)
# cell = (n layers * n directions, batch size, hid dim)
# rnn的输出总是来自顶部的隐藏层
return hidden, cell
class Decoder(nn.Module):
def __init__(self, output_dim, emb_dim, hid_dim, n_layers, dropout):
super().__init__()
self.output_dim = output_dim
self.hid_dim = hid_dim
self.n_layers = n_layers
self.embedding = nn.Embedding(output_dim, emb_dim)
self.rnn = nn.LSTM(emb_dim, hid_dim, n_layers, dropout=dropout)
self.fc_out = nn.Linear(hid_dim, output_dim)
self.dropout = nn.Dropout(dropout)
def forward(self, input, hidden, cell):
# 各输入的形状
# input = (batch size)
# hidden = (n layers * n directions, batch size, hid dim)
# cell = (n layers * n directions, batch size, hid dim)
# LSTM是单向的 ==> n directions == 1
# hidden = (n layers, batch size, hid dim)
# cell = (n layers, batch size, hid dim)
input = input.unsqueeze(0) # (batch size) --> [1, batch size)
embedded = self.dropout(self.embedding(input)) # (1, batch size, emb dim)
output, (hidden, cell) = self.rnn(embedded, (hidden, cell))
# LSTM理论上的输出形状
# output = (seq len, batch size, hid dim * n directions)
# hidden = (n layers * n directions, batch size, hid dim)
# cell = (n layers * n directions, batch size, hid dim)
# 解码器中的序列长度 seq len == 1
# 解码器的LSTM是单向的 n directions == 1 则实际上
# output = (1, batch size, hid dim)
# hidden = (n layers, batch size, hid dim)
# cell = (n layers, batch size, hid dim)
prediction = self.fc_out(output.squeeze(0))
# prediction = (batch size, output dim)
return prediction, hidden, cell
class Seq2Seq(nn.Module):
def __init__(self, input_word_count, output_word_count, encode_dim, decode_dim, hidden_dim, n_layers,
encode_dropout, decode_dropout, device):
"""
:param input_word_count: 英文词表的长度 34737
:param output_word_count: 中文词表的长度 4015
:param encode_dim: 编码器的词嵌入维度
:param decode_dim: 解码器的词嵌入维度
:param hidden_dim: LSTM的隐藏层维度
:param n_layers: 采用n层LSTM
:param encode_dropout: 编码器的dropout概率
:param decode_dropout: 编码器的dropout概率
:param device: cuda / cpu
"""
super().__init__()
self.encoder = Encoder(input_word_count, encode_dim, hidden_dim, n_layers, encode_dropout)
self.decoder = Decoder(output_word_count, decode_dim, hidden_dim, n_layers, decode_dropout)
self.device = device
def forward(self, src):
# src = (src len, batch size)
# 编码器的隐藏层输出将作为解码器的第一个隐藏层输入
hidden, cell = self.encoder(src)
# 解码器的第一个输入应该是起始标识符模型预测完整代码
提示：预测代码是我们基于训练及验证代码进行改造的，不一定完全正确，可以参考后自行修改~
import torch
import torch.nn as nn
import pickle
class Encoder(nn.Module):
def __init__(self, input_dim, emb_dim, hid_dim, n_layers, dropout):
super().__init__()
self.hid_dim = hid_dim
self.n_layers = n_layers
self.embedding = nn.Embedding(input_dim, emb_dim) # 词嵌入
self.rnn = nn.LSTM(emb_dim, hid_dim, n_layers, dropout=dropout)
self.dropout = nn.Dropout(dropout)
def forward(self, src):
# src = (src len, batch size)
embedded = self.dropout(self.embedding(src))
# embedded = (src len, batch size, emb dim)
outputs, (hidden, cell) = self.rnn(embedded)
# outputs = (src len, batch size, hid dim * n directions)
# hidden = (n layers * n directions, batch size, hid dim)
# cell = (n layers * n directions, batch size, hid dim)
# rnn的输出总是来自顶部的隐藏层
return hidden, cell
class Decoder(nn.Module):
def __init__(self, output_dim, emb_dim, hid_dim, n_layers, dropout):
super().__init__()
self.output_dim = output_dim
self.hid_dim = hid_dim
self.n_layers = n_layers
self.embedding = nn.Embedding(output_dim, emb_dim)
self.rnn = nn.LSTM(emb_dim, hid_dim, n_layers, dropout=dropout)
self.fc_out = nn.Linear(hid_dim, output_dim)
self.dropout = nn.Dropout(dropout)
def forward(self, input, hidden, cell):
# 各输入的形状
# input = (batch size)
# hidden = (n layers * n directions, batch size, hid dim)
# cell = (n layers * n directions, batch size, hid dim)
# LSTM是单向的 ==> n directions == 1
# hidden = (n layers, batch size, hid dim)
# cell = (n layers, batch size, hid dim)
input = input.unsqueeze(0) # (batch size) --> [1, batch size)
embedded = self.dropout(self.embedding(input)) # (1, batch size, emb dim)
output, (hidden, cell) = self.rnn(embedded, (hidden, cell))
# LSTM理论上的输出形状
# output = (seq len, batch size, hid dim * n directions)
# hidden = (n layers * n directions, batch size, hid dim)
# cell = (n layers * n directions, batch size, hid dim)
# 解码器中的序列长度 seq len == 1
# 解码器的LSTM是单向的 n directions == 1 则实际上
# output = (1, batch size, hid dim)
# hidden = (n layers, batch size, hid dim)
# cell = (n layers, batch size, hid dim)
prediction = self.fc_out(output.squeeze(0))
# prediction = (batch size, output dim)
return prediction, hidden, cell
class Seq2Seq(nn.Module):
def __init__(self, input_word_count, output_word_count, encode_dim, decode_dim, hidden_dim, n_layers,
encode_dropout, decode_dropout, device):
"""
:param input_word_count: 英文词表的长度 34737
:param output_word_count: 中文词表的长度 4015
:param encode_dim: 编码器的词嵌入维度
:param decode_dim: 解码器的词嵌入维度
:param hidden_dim: LSTM的隐藏层维度
:param n_layers: 采用n层LSTM
:param encode_dropout: 编码器的dropout概率
:param decode_dropout: 编码器的dropout概率
:param device: cuda / cpu
"""
super().__init__()
self.encoder = Encoder(input_word_count, encode_dim, hidden_dim, n_layers, encode_dropout)
self.decoder = Decoder(output_word_count, decode_dim, hidden_dim, n_layers, decode_dropout)
self.device = device
def forward(self, src):
# src = (src len, batch size)
# 编码器的隐藏层输出将作为解码器的第一个隐藏层输入
hidden, cell = self.encoder(src)
# 解码器的第一个输入应该是起始标识符结束语
- 亲爱的读者，感谢您花时间阅读我们的博客。我们非常重视您的反馈和意见，因此在这里鼓励您对我们的博客进行评论。
- 您的建议和看法对我们来说非常重要，这有助于我们更好地了解您的需求，并提供更高质量的内容和服务。
- 无论您是喜欢我们的博客还是对其有任何疑问或建议，我们都非常期待您的留言。让我们一起互动，共同进步！谢谢您的支持和参与！
- 我会坚持不懈地创作，并持续优化博文质量，为您提供更好的阅读体验。
- 谢谢您的阅读！