本文为转载，原文链接: https://wmathor.com/index.php/archives/1451/
本文主要介绍一下如何使用 PyTorch 复现 Seq2Seq (with Attention)，实现简单的机器翻译任务，请先阅读论文 Neural Machine Translation by Jointly Learning to Align and Translate，之后花上 15 分钟阅读我的这两篇文章 Seq2Seq 与注意力机制，图解 Attention，最后再来看文本，方能达到醍醐灌顶，事半功倍的效果。
数据预处理
数据预处理的代码其实就是调用各种 API，我不希望读者被这些不太重要的部分分散了注意力，因此这里我不贴代码，仅口述一下带过即可。
这里采用的数据集是torchtext的multi30k
如下图所示，本文使用的是德语→英语数据集，输入是德语，并且输入的每个句子开头和结尾都带有特殊的标识符。输出是英语，并且输出的每个句子开头和结尾也都带有特殊标识符
不管是英语还是德语，每句话长度都是不固定的，所以我对于每个 batch 内的句子，将它们的长度通过加 [seq_len, batch_size]
随便打印一条数据，看一下数据封装的形式
在数据预处理的时候，需要将源句子和目标句子分开构建字典，也就是单独对德语构建一个词库，对英语构建一个词库
Encoder
Encoder这里使用的是单层双向GRU
双向GRU的隐藏状态输出由两个向量拼接而成，例如,……
所有时刻的最后一层隐藏状态就构成了 GRU 的 output.
假设这是个m层GRU，那么最后一个时刻所有层中的隐藏状态就构成了GRU的final hidden states.
其中
所以，
根据论文，或者你看了我的图解 Attention 这篇文章就会知道，我们需要的是 hidden 的最后一层输出（包括正向和反向），因此我们可以通过 hidden[-2,:,:] 和 hidden[-1,:,:] 取出最后一层的 hidden states，将它们拼接起来记作
最后一个细节之处在于， 的维度是 [batch_size, en_hid_dim*2]，即便是没有 Attention 机制，将 作为 Decoder 的初始隐藏状态也不对，因为维度不匹配，Decoder 的初始隐藏状态是三维的，而现在我们的 是二维的，因此需要将 的维度转为三维，并且还要调整各个维度上的值。首先我通过一个全连接神经网络，将 的维度变为 [batch_size, dec_hid_dim]
Encoder 的细节就这么多，下面直接上代码，我的代码风格是，注释在上，代码在下
class Encoder(nn.Module):
def __init__(self, input_dim, emb_dim, enc_hid_dim, dec_hid_dim, dropout):
super().__init__()
self.embedding = nn.Embedding(input_dim, emb_dim)
# single layer, bi-direction GRU
self.rnn = nn.GRU(emb_dim, enc_hid_dim, bidirectional=True)
self.fc = nn.Linear(enc_hid_dim * 2, dec_hid_dim)
self.dropout = nn.Dropout(dropout)
def forward(self, src):
'''
:param src: [src_len, batch_size]
:return:
'''
src = src.transpose(0, 1) # src = [batch_size, src_len]
# embedded = [src_len, batch_size, emb_dim]
embedded = self.dropout(self.embedding(src)).transpose(0, 1)
# enc_output = [src_len, batch_size, hid_dim*num_directions]
# enc_hidden = [n_layers * num_directions, batch_size, hid_dim]
enc_output, enc_hidden = self.rnn(embedded) # if h_0 is not given, it will be set 0 acquiescently
# enc_hidden is stacked [forward_1, backward_1, forward_2, backward_2, ...]
# enc_output are always from the last layer
# enc_hidden [-2, :, : ] is the last of the forwards RNN
# enc_hidden [-1, :, : ] is the last of the backwards RNN
# initial decoder hidden is final hidden state of the forwards and backwards
# encoder RNNs fed through a linear layer
# concatenate the hidden_state of last two layers
# s = [batch_size, dec_hid_dim]
s = torch.tanh(self.fc(torch.cat((enc_hidden[-2, :, :], enc_hidden[-1, :, :]), dim=1)))
return enc_output, s
Attention
attention 无非就是三个公式
其中指的就是Encoder中的变量，就是指的Encoder中的变量enc_output，其实就是一个简单的全连接神经网络。
我们可以从最后一个公式反推各个变量的维度是什么，或者维度有什么要求
首先 的维度应该是 [batch_size, src_len]，这是毋庸置疑的，那么 的维度也应该是 [batch_size, src_len]，或者 是个三维的，但是某个维度值为 1，可以通过 squeeze() 变成两维的。这里我们先假设 的维度是 [batch_size, src_len, 1]，等会儿我再解释为什么要这样假设
继续往上推，变量 的维度就应该是 [?, 1]，? 表示我暂时不知道它的值应该是多少。 的维度应该是 [batch_size, src_len, ?]
现在已知 的维度是 [batch_size, src_len, enc_hid_dim*2]， 目前的维度是 [batch_size, dec_hid_dim]，这两个变量需要做拼接，送入全连接神经网络，因此我们首先需要将 的维度变成 [batch_size, src_len, dec_hid_dim]，拼接之后的维度就变成 [batch_size, src_len, enc_hid_dim*2+dec_hid_dim]，于是 这个函数的输入输出值也就有了
attn = nn.Linear(enc_hid_dim*2+dec_hid_dim, ?)
到此为止，除了? 部分的值不清楚，其它所有维度都推导出来了。现在我们回过头思考一下 ? 设置成多少，好像其实并没有任何限制，所以我们可以设置 ? 为任何值（在代码中我设置 ? 为 dec_hid_dim）
Attention细节就这么多，下面给出代码
class Attention(nn.Module):
def __init__(self, enc_hid_dim, dec_hid_dim):
super().__init__()
# [size(h_t)+size(s_{t-1}), dec_hid_dim]
self.attn = nn.Linear((enc_hid_dim * 2) + dec_hid_dim, dec_hid_dim, bias=False)
self.v = nn.Linear(dec_hid_dim, 1, bias=False)
def forward(self, s, enc_output):
# s = [batch_size, dec_hid_dim]
# enc_output = [src_len, batch_size, enc_hid_dim * 2]
batch_size = enc_output.shape[1]
src_len = enc_output.shape[0]
# repeat decoder hidden state src_len times
# s = [batch_size, src_len, enc_hid_dim * 2]
# enc_output = [batch_size, src_len, enc_hid_dim * 2]
s = s.unsqueeze(1).repeat(1, src_len, 1)
enc_output = enc_output.transpose(0, 1)
# energy = [batch_size, src_len, dec_hid_dim]
energy = torch.tanh(self.attn(torch.cat((s, enc_output), dim=2)))
# attention = [batch_size, src_len]
attention = self.v(energy).squeeze(2)
return F.softmax(attention, dim=1)
Seq2Seq(with Attention)
我调换一下顺序，先讲 Seq2Seq，再讲 Decoder 的部分
传统 Seq2Seq 是直接将句子中每个词连续不断输入 Decoder 进行训练，而引入 Attention 机制之后，我需要能够人为控制一个词一个词进行输入（因为输入每个词到 Decoder，需要再做一些运算），所以在代码中会看到我使用了 for 循环，循环 trg_len-1 次（开头的
并且训练过程中我使用了一种叫做 Teacher Forcing 的机制，保证训练速度的同时增加鲁棒性，如果不了解 Teacher Forcing 可以看我的这篇文章
思考一下 for 循环中应该要做哪些事？首先要将变量传入 Decoder，由于 Attention 的计算是在 Decoder 的内部进行的，所以我需要将 dec_input、s、enc_output 这三个变量传入 Decoder，Decoder 会返回 dec_output 以及新的 s。之后根据概率对 dec_output 做 Teacher Forcing 即可
Seq2Seq 细节就这么多，下面给出代码
class Seq2Seq(nn.Module):
def __init__(self, encoder, decoder, device):
super().__init__()
self.encoder = encoder
self.decoder = decoder
self.device = device
def forward(self, src, trg, teacher_forcing_ratio=0.5):
# src = [src_len, batch_size]
# trg = [trg_len, batch_size]
# teacher_forcing_ratio is probability to use teacher forcing (scheduled sampling)
batch_size = src.shape[1]
trg_len = trg.shape[0]
trg_vocab_size = self.decoder.output_dim
# tensor to store decoder outputs
outputs = torch.zeros(trg_len, batch_size, trg_vocab_size).to(self.device)
# enc_output is all hidden states of the input sequence, back and forwards
# s is the final forward and backward hidden states, passed through a linear layer
# enc_output : [src_len, batch_size, enc_hid_dim * 2]
# s : [batch_size, dec_hid_dim]
enc_output, s = self.encoder(src)
# first input to the decoder is the tokens
dec_input = trg[0, :]
for t in range(1, trg_len):
# insert dec_input token embedding, previous hidden state and all encoder hidden states
# receive output tensor (predictions) and new hidden state
dec_output, s = self.decoder(dec_input, s, enc_output)
# place predictions in a tensor holding predictions for each token
outputs[t] = dec_output
# decide if we are going to use teacher forcing or not
teacher_force = random.random() < teacher_forcing_ratio
# get the highest predicted token from our predictions
top1 = dec_output.argmax(1)
# if teacher forcing, use actural next token as input
# if not, use predicted token
dec_input = trg[t] if teacher_force else top1
return outputs
Decoder
Decoder这里使用单向单层GRU
Decoder 部分实际上也就是三个公式
指的是 Encoder 中的变量 enc_output， 指的是将 dec_input 经过 WordEmbedding 后得到的结果， 函数实际上就是为了转换维度，因为需要的输出是 TRG_VOCAB_SIZE 大小。其中有个细节，GRU 的参数只有两个，一个输入，一个隐藏层输入，但是上面的公式有三个变量，所以我们应该选一个作为隐藏层输入，另外两个 "整合" 一下，作为输入
我们从第一个公式正推各个变量的维度是什么
首先在 Encoder 中最开始先调用一次 Attention，得到权重 ，它的维度是 [batch_size, src_len]，而 的维度是 [src_len, batch_size, enc_hid_dim*2]，它俩要相乘，同时应该保留 batch_size 这个维度，所以应该先将 扩展一维，然后调换一下维度的顺序，之后再按照 batch 相乘（即同一个 batch 内的矩阵相乘）
a = a.unsqueeze(1) # [batch_size, 1, src_len]
H = H.transpose(0, 1) # [batch_size, src_len, enc_hid_dim*2]
c = torch.bmm(a, h) # [batch_size, 1, enc_hid_dim*2]
前面也说了，由于 GRU 不需要三个变量，所以需要将和 整合一下， 实际上就是 Seq2Seq 类中的 变量，它的维度是 [batch_size]，因此先将 扩展一个维度，再通过 WordEmbedding，这样他就变成 [batch_size, 1, emb_dim]。最后对 和 进行 concat
y = y.unsqueeze(1) # [batch_size, 1]
emb_y = self.emb(y) # [batch_size, 1, emb_dim]
rnn_input = torch.cat((emb_y, c), dim=2) # [batch_size, 1, emb_dim+enc_hid_dim*2]
的维度是 [batch_size, dec_hid_dim]，所以应该先将其拓展一个维度 (layer*num_direction维)
rnn_input = rnn_input.transpose(0, 1) # [1, batch_size, emb_dim+enc_hid_dim*2]
s = s.unsqueeze(0) # [1, batch_size, dec_hid_dim]
# dec_output = [1, batch_size, dec_hid_dim]
# dec_hidden = [1, batch_size, dec_hid_dim] = s (new, is not s previously)
dec_output, dec_hidden = self.rnn(rnn_input, s)
最后一个公式，需要将三个变量全部拼接在一起，然后通过一个全连接神经网络，得到最终的预测。我们先分析下这个三个变量的维度， 的维度是 [batch_size, 1, emb_dim]， 的维度是 [batch_size, 1, enc_hid_dim]， 的维度是 [1, batch_size, dec_hid_dim]，因此我们可以像下面这样把他们全部拼接起来
emd_y = emb_y.squeeze(1) # [batch_size, emb_dim]
c = w.squeeze(1) # [batch_size, enc_hid_dim*2]
s = s.squeeze(0) # [batch_size, dec_hid_dim]
fc_input = torch.cat((emb_y, c, s), dim=1) # [batch_size, enc_hid_dim*2+dec_hid_dim+emb_hid]
以上就是 Decoder 部分的细节，下面给出代码（上面的那些只是示例代码，和下面代码变量名可能不一样）
class Decoder(nn.Module):
def __init__(self, output_dim, emb_dim, enc_hid_dim, dec_hid_dim, dropout, attention):
super().__init__()
self.output_dim = output_dim
self.attention = attention
self.embedding = nn.Embedding(output_dim, emb_dim)
self.rnn = nn.GRU(enc_hid_dim * 2 + emb_dim, dec_hid_dim)
self.fc_out = nn.Linear(enc_hid_dim * 2 + dec_hid_dim + emb_dim, output_dim)
self.dropout = nn.Dropout(dropout)
def forward(self, dec_input, s, enc_output):
# dec_input = [batch_size]
# s = [batch_size, dec_hid_dim]
# enc_output = [src_len, batch_size, enc_hid_dim *2]
# dec_input = [batch_size,1]
dec_input = dec_input.unsqueeze(1)
# embedded = [1, batch_size, emb_dim]
embedded = self.dropout(self.embedding(dec_input)).transpose(0, 1)
# s = [batch_size, dec_hid_dim]
# enc_output = [src_len, batch_size, enc_hid_dim *2]
# a = [batch_size, 1, src_len]
a = self.attention(s, enc_output).unsqueeze(1)
# enc_output = [batch_size, src_len, enc_hid_dim * 2]
enc_output = enc_output.transpose(0, 1)
# c = [1, batch_size, enc_hid_dim * 2]
c = torch.bmm(a, enc_output).transpose(0, 1)
# torch.bmm: Performs a batch matrix-matrix product of matrices stored in input and mat2
# rnn_input = [1, batch_size, (enc_hid_dim*2) + emb_dim]
rnn_input = torch.cat((embedded, c), dim=2)
# dec_output = [src_len(=1), batch_size, dec_hid_dim]
# dec_hidden = [n_layers*num_directions, batch_size, dec_hid_dim]
dec_output, dec_hidden = self.rnn(rnn_input, s.unsqueeze(0))
# embedded = [batch_size, emb_dim]
# dec_output = [batch_size, dec_hid_dim]
# c = [batch_size, enc_hid_dim * 2]
embedded = embedded.squeeze(0)
dec_output = dec_output.squeeze(0)
c = c.squeeze(0)
# pred = [batch_size, output_dim]
pred = self.fc_out(torch.cat((dec_output, c, embedded), dim=1))
return pred, dec_hidden.squeeze(0)
定义模型
INPUT_DIM = len(SRC.vocab)
OUTPUT_DIM = len(TRG.vocab)
ENC_EMB_DIM = 256
DEC_EMB_DIM = 256
ENC_HID_DIM = 512
DEC_HID_DIM = 512
ENC_DROPOUT = 0.5
DEC_DROPOUT = 0.5
attn = Attention(ENC_HID_DIM, DEC_HID_DIM)
enc = Encoder(INPUT_DIM, ENC_EMB_DIM, ENC_HID_DIM, DEC_HID_DIM, ENC_DROPOUT)
dec = Decoder(OUTPUT_DIM, DEC_EMB_DIM, ENC_HID_DIM, DEC_HID_DIM, DEC_DROPOUT, attn)
model = Seq2Seq(enc, dec, device).to(device)
TRG_PAD_IDX = TRG.vocab.stoi[TRG.pad_token]
criterion = nn.CrossEntropyLoss(ignore_index = TRG_PAD_IDX).to(device)
optimizer = optim.Adam(model.parameters(), lr=1e-3)
倒数第二行CrossEntropyLoss() 中的参数很少见，ignore_index=TRG_PAD_IDX，这个参数的作用是忽略某一类别，不计算其 loss，但是要注意，忽略的是真实值中的类别，例如下面的代码，真实值的类别都是 1，而预测值全部预测认为是 2（下标从 0 开始），同时 loss function 设置忽略第一类的 loss，此时会打印出 0
label = torch.tensor([1, 1, 1])
pred = torch.tensor([[0.1, 0.2, 0.6], [0.2, 0.1, 0.8], [0.1, 0.1, 0.9]])
loss_fn = nn.CrossEntropyLoss(ignore_index=1)
print(loss_fn(pred, label).item()) # 0
如果设置 loss function 忽略第二类，此时 loss 并不会为 0
label = torch.tensor([1, 1, 1])
pred = torch.tensor([[0.1, 0.2, 0.6], [0.2, 0.1, 0.8], [0.1, 0.1, 0.9]])
loss_fn = nn.CrossEntropyLoss(ignore_index=2)
print(loss_fn(pred, label).item()) # 1.359844
完整代码
#!/usr/bin/env python
# coding: utf-8
# ### Preparing Data
#
# In[1]:
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torchtext.legacy.datasets import Multi30k
from torchtext.legacy.data import Field, BucketIterator
import spacy
import numpy as np
import random
import math
import time
# Set the random seeds for reproducability
#
#
# In[2]:
SEED = 1234
random.seed(SEED)
np.random.seed(SEED)
torch.manual_seed(SEED)
torch.cuda.manual_seed(SEED)
# 加载German 和 English spaCy模型
# In[3]:
spacy_de = spacy.load('de_core_news_sm')
spacy_en = spacy.load('en_core_web_sm')
# 创建tokenizer
# In[4]:
def tokenize_de(text):
# Tokenizes German text from a string into a list of strings
return [tok.text for tok in spacy_de.tokenizer(text)]
def tokenize_en(text):
# Tokenizes English text from a string into a list of strings
return [tok.text for tok in spacy_en.tokenizer(text)]
# 创建Field对象（torchtext的一种数据类型）
# Field类将普通文本转为tensor
# see: https://torchtext.readthedocs.io/en/latest/data.html#field
# In[5]:
SRC = Field(tokenize=tokenize_de,
init_token='',
eos_token='',
lower=True)
TRG = Field(tokenize=tokenize_en,
init_token='',
eos_token='',
lower=True)
# 加载数据
# In[7]:
# Create dataset objects for splits of the Multi30k dataset.
train_data, valid_data, test_data = Multi30k.splits(exts=('.de', '.en'), fields=(SRC, TRG))
# 建立词表
# In[8]:
SRC.build_vocab(train_data, min_freq=2) #min_freq:最小频率
TRG.build_vocab(train_data, min_freq=2)
# 定义设备
# In[6]:
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# 创建迭代器
# In[9]:
BATCH_SIZE = 128
train_iterator, valid_iterator, test_iterator = BucketIterator.splits(
(train_data, valid_data, test_data),
batch_size=BATCH_SIZE,
device=device)
# 查看数据size
#
# 可以看到经过padding之后，一个batch内的句子，长度都是相同的，不同的batch内的句子长度不一定相同。
# In[11]:
for i, it in enumerate(iter(train_iterator)):
if i > 10:
break
src = it.src # German
trg = it.trg # English
print(src.shape, trg.shape)
# torch.Size([seq_len,batch_size])
# 查看数据形式
# In[10]:
batch_idx = 0
data = next(iter(train_iterator))
for idx in data.src[:, batch_idx].cpu().numpy():
print(SRC.vocab.itos[idx], end=' ')
print()
for idx in data.trg[:, batch_idx].cpu().numpy():
print(TRG.vocab.itos[idx], end=' ')
# 创建 Seq2Seq Model
#
# 这里Encoder采用单层双向GRU
# In[12]:
class Encoder(nn.Module):
def __init__(self, input_dim, emb_dim, enc_hid_dim, dec_hid_dim, dropout):
super().__init__()
self.embedding = nn.Embedding(input_dim, emb_dim)
# single layer, bi-direction GRU
self.rnn = nn.GRU(emb_dim, enc_hid_dim, bidirectional=True)
self.fc = nn.Linear(enc_hid_dim * 2, dec_hid_dim)
self.dropout = nn.Dropout(dropout)
def forward(self, src):
'''
:param src: [src_len, batch_size]
:return:
'''
src = src.transpose(0, 1) # src = [batch_size, src_len]
# embedded = [src_len, batch_size, emb_dim]
embedded = self.dropout(self.embedding(src)).transpose(0, 1)
# enc_output = [src_len, batch_size, hid_dim*num_directions]
# enc_hidden = [n_layers * num_directions, batch_size, hid_dim]
enc_output, enc_hidden = self.rnn(embedded) # if h_0 is not given, it will be set 0 acquiescently
# enc_hidden is stacked [forward_1, backward_1, forward_2, backward_2, ...]
# enc_output are always from the last layer
# enc_hidden [-2, :, : ] is the last of the forwards RNN
# enc_hidden [-1, :, : ] is the last of the backwards RNN
# initial decoder hidden is final hidden state of the forwards and backwards
# encoder RNNs fed through a linear layer
# concatenate the hidden_state of last two layers
# s = [batch_size, dec_hid_dim]
s = torch.tanh(self.fc(torch.cat((enc_hidden[-2, :, :], enc_hidden[-1, :, :]), dim=1)))
return enc_output, s
# In[13]:
class Attention(nn.Module):
def __init__(self, enc_hid_dim, dec_hid_dim):
super().__init__()
# [size(h_t)+size(s_{t-1}), dec_hid_dim]
self.attn = nn.Linear((enc_hid_dim * 2) + dec_hid_dim, dec_hid_dim, bias=False)
self.v = nn.Linear(dec_hid_dim, 1, bias=False)
def forward(self, s, enc_output):
# s = [batch_size, dec_hid_dim]
# enc_output = [src_len, batch_size, enc_hid_dim * 2]
batch_size = enc_output.shape[1]
src_len = enc_output.shape[0]
# repeat decoder hidden state src_len times
# s = [batch_size, src_len, enc_hid_dim * 2]
# enc_output = [batch_size, src_len, enc_hid_dim * 2]
s = s.unsqueeze(1).repeat(1, src_len, 1)
enc_output = enc_output.transpose(0, 1)
# energy = [batch_size, src_len, dec_hid_dim]
energy = torch.tanh(self.attn(torch.cat((s, enc_output), dim=2)))
# attention = [batch_size, src_len]
attention = self.v(energy).squeeze(2)
return F.softmax(attention, dim=1)
# Seq2Seq Model
# In[14]:
class Seq2Seq(nn.Module):
def __init__(self, encoder, decoder, device):
super().__init__()
self.encoder = encoder
self.decoder = decoder
self.device = device
def forward(self, src, trg, teacher_forcing_ratio=0.5):
# src = [src_len, batch_size]
# trg = [trg_len, batch_size]
# teacher_forcing_ratio is probability to use teacher forcing (scheduled sampling)
batch_size = src.shape[1]
trg_len = trg.shape[0]
trg_vocab_size = self.decoder.output_dim
# tensor to store decoder outputs
outputs = torch.zeros(trg_len, batch_size, trg_vocab_size).to(self.device)
# enc_output is all hidden states of the input sequence, back and forwards
# s is the final forward and backward hidden states, passed through a linear layer
# enc_output : [src_len, batch_size, enc_hid_dim * 2]
# s : [batch_size, dec_hid_dim]
enc_output, s = self.encoder(src)
# first input to the decoder is the tokens
dec_input = trg[0, :]
for t in range(1, trg_len):
# insert dec_input token embedding, previous hidden state and all encoder hidden states
# receive output tensor (predictions) and new hidden state
dec_output, s = self.decoder(dec_input, s, enc_output)
# place predictions in a tensor holding predictions for each token
outputs[t] = dec_output
# decide if we are going to use teacher forcing or not
teacher_force = random.random() < teacher_forcing_ratio
# get the highest predicted token from our predictions
top1 = dec_output.argmax(1)
# if teacher forcing, use actural next token as input
# if not, use predicted token
dec_input = trg[t] if teacher_force else top1
return outputs
# Decoder
# In[15]:
class Decoder(nn.Module):
def __init__(self, output_dim, emb_dim, enc_hid_dim, dec_hid_dim, dropout, attention):
super().__init__()
self.output_dim = output_dim
self.attention = attention
self.embedding = nn.Embedding(output_dim, emb_dim)
self.rnn = nn.GRU(enc_hid_dim * 2 + emb_dim, dec_hid_dim)
self.fc_out = nn.Linear(enc_hid_dim * 2 + dec_hid_dim + emb_dim, output_dim)
self.dropout = nn.Dropout(dropout)
def forward(self, dec_input, s, enc_output):
# dec_input = [batch_size]
# s = [batch_size, dec_hid_dim]
# enc_output = [src_len, batch_size, enc_hid_dim *2]
# dec_input = [batch_size,1]
dec_input = dec_input.unsqueeze(1)
# embedded = [1, batch_size, emb_dim]
embedded = self.dropout(self.embedding(dec_input)).transpose(0, 1)
# s = [batch_size, dec_hid_dim]
# enc_output = [src_len, batch_size, enc_hid_dim *2]
# a = [batch_size, 1, src_len]
a = self.attention(s, enc_output).unsqueeze(1)
# enc_output = [batch_size, src_len, enc_hid_dim * 2]
enc_output = enc_output.transpose(0, 1)
# c = [1, batch_size, enc_hid_dim * 2]
c = torch.bmm(a, enc_output).transpose(0, 1)
# torch.bmm: Performs a batch matrix-matrix product of matrices stored in input and mat2
# rnn_input = [1, batch_size, (enc_hid_dim*2) + emb_dim]
rnn_input = torch.cat((embedded, c), dim=2)
# dec_output = [src_len(=1), batch_size, dec_hid_dim]
# dec_hidden = [n_layers*num_directions, batch_size, dec_hid_dim]
dec_output, dec_hidden = self.rnn(rnn_input, s.unsqueeze(0))
# embedded = [batch_size, emb_dim]
# dec_output = [batch_size, dec_hid_dim]
# c = [batch_size, enc_hid_dim * 2]
embedded = embedded.squeeze(0)
dec_output = dec_output.squeeze(0)
c = c.squeeze(0)
# pred = [batch_size, output_dim]
pred = self.fc_out(torch.cat((dec_output, c, embedded), dim=1))
return pred, dec_hidden.squeeze(0)
# 定义模型
# In[16]:
INPUT_DIM = len(SRC.vocab)
OUTPUT_DIM = len(TRG.vocab)
ENC_EMB_DIM = 256
DEC_EMB_DIM = 256
ENC_HID_DIM = 512
DEC_HID_DIM = 512
ENC_DROPOUT = 0.5
DEC_DROPOUT = 0.5
attn = Attention(ENC_HID_DIM, DEC_HID_DIM)
enc = Encoder(INPUT_DIM, ENC_EMB_DIM, ENC_HID_DIM, DEC_HID_DIM, ENC_DROPOUT)
dec = Decoder(OUTPUT_DIM, DEC_EMB_DIM, ENC_HID_DIM, DEC_HID_DIM, DEC_DROPOUT, attn)
model = Seq2Seq(enc, dec, device).to(device)
TRG_PAD_IDX = TRG.vocab.stoi[TRG.pad_token]
criterion = nn.CrossEntropyLoss(ignore_index=TRG_PAD_IDX).to(device)
optimizer = optim.Adam(model.parameters(), lr=1e-3)
# 定义训练函数
# In[18]:
def train(model, iterator, optimizer, criterion):
model.train()
epoch_loss = 0
for i, batch in enumerate(iterator):
src = batch.src
trg = batch.trg # [trg_len, batch_size]
# pred = [trg_len, batch_size, pred_dim]
pred = model(src, trg)
pred_dim = pred.shape[-1]
# trg = [(trg_len - 1)*batch_size]
# pred = [(trg_len - 1)*batch_size]
trg = trg[1:].view(-1)
pred = pred[1:].view(-1, pred_dim)
loss = criterion(pred, trg)
optimizer.zero_grad()
loss.backward()
optimizer.step()
epoch_loss += loss.item()
return epoch_loss / len(iterator)
# 定义评估函数
# In[19]:
def evaluate(model, iterator, criterion):
model.eval()
epoch_loss = 0
with torch.no_grad():
for i, batch in enumerate(iterator):
src = batch.src
trg = batch.trg # trg = [trg_len, batch_size]
# output = [trg_len, batch_size, output_dim]
output = model(src, trg, 0) # turn off teacher forcing
output_dim = output.shape[-1]
# trg = [(trg_len - 1) * batch_size]
# output = [(trg_len - 1) * batch_size, output_dim]
output = output[1:].view(-1, output_dim)
trg = trg[1:].view(-1)
loss = criterion(output, trg)
epoch_loss += loss.item()
return epoch_loss / len(iterator)
# 定义一个时间函数
#
# In[20]:
def epoch_time(start_time, end_time):
elapsed_time = end_time - start_time
elapsed_mins = int(elapsed_time / 60)
elapsed_secs = int(elapsed_time - (elapsed_mins * 60))
return elapsed_mins, elapsed_secs
# 训练
# In[21]:
best_valid_loss = float('inf')
for epoch in range(10):
start_time = time.time()
train_loss = train(model, train_iterator, optimizer, criterion)
valid_loss = evaluate(model, valid_iterator, criterion)
end_time = time.time()
epoch_mins, epoch_secs = epoch_time(start_time, end_time)
if valid_loss < best_valid_loss:
best_valid_loss = valid_loss
torch.save(model.state_dict(), 'tut3-model.pt')
print(f'Epoch: {epoch+1:02} | Time: {epoch_mins}m {epoch_secs}s')
print(f'\tTrain Loss: {train_loss:.3f} | Train PPL: {math.exp(train_loss):7.3f}')
print(f'\t Val. Loss: {valid_loss:.3f} | Val. PPL: {math.exp(valid_loss):7.3f}')
# 保存模型与测试
# In[ ]:
model.load_state_dict(torch.load('tut3-model.pt'))
test_loss = evaluate(model, test_iterator, criterion)
print(f'| Test Loss: {test_loss:.3f} | Test PPL: {math.exp(test_loss):7.3f} |')