第N11周：seq2seq翻译实战-Pytorch复现

任务：
●为解码器添加上注意力机制

一、前期准备工作

from __future__ import unicode_literals, print_function, division
from io import open
import unicodedata
import string
import re
import random

import torch
import torch.nn as nn
from torch import optim
import torch.nn.functional as F

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

代码输出

cpu

1. 搭建语言类

SOS_token = 0
EOS_token = 1

# 语言类，方便对语料库进行操作
class Lang:
    def __init__(self, name):
        self.name = name
        self.word2index = {}
        self.word2count = {}
        self.index2word = {0: "SOS", 1: "EOS"}
        self.n_words    = 2  # Count SOS and EOS

    def addSentence(self, sentence):
        for word in sentence.split(' '):
            self.addWord(word)

    def addWord(self, word):
        if word not in self.word2index:
            self.word2index[word] = self.n_words
            self.word2count[word] = 1
            self.index2word[self.n_words] = word
            self.n_words += 1
        else:
            self.word2count[word] += 1

2. 文本处理函数

def unicodeToAscii(s):
    return ''.join(
        c for c in unicodedata.normalize('NFD', s)
        if unicodedata.category(c) != 'Mn'
    )

# 小写化，剔除标点与非字母符号
def normalizeString(s):
    s = unicodeToAscii(s.lower().strip())
    s = re.sub(r"([.!?])", r" \1", s)
    s = re.sub(r"[^a-zA-Z.!?]+", r" ", s)
    return s

3. 文件读取函数

def readLangs(lang1, lang2, reverse=False):
    print("Reading lines...")

    # 以行为单位读取文件
    lines = open('N11/%s-%s.txt'%(lang1,lang2), encoding='utf-8').\
            read().strip().split('\n')

    # 将每一行放入一个列表中
    # 一个列表中有两个元素，A语言文本与B语言文本
    pairs = [[normalizeString(s) for s in l.split('\t')] for l in lines]

    # 创建Lang实例，并确认是否反转语言顺序
    if reverse:
        pairs       = [list(reversed(p)) for p in pairs]
        input_lang  = Lang(lang2)
        output_lang = Lang(lang1)
    else:
        input_lang  = Lang(lang1)
        output_lang = Lang(lang2)

    return input_lang, output_lang, pairs

.startswith(eng_prefixes) 是字符串方法 startswith() 的调用。它用于检查一个字符串是否以指定的前缀开始。

MAX_LENGTH = 10      # 定义语料最长长度

eng_prefixes = (
    "i am ", "i m ",
    "he is", "he s ",
    "she is", "she s ",
    "you are", "you re ",
    "we are", "we re ",
    "they are", "they re "
)

def filterPair(p):
    return len(p[0].split(' ')) < MAX_LENGTH and \
           len(p[1].split(' ')) < MAX_LENGTH and p[1].startswith(eng_prefixes)

def filterPairs(pairs):
    # 选取仅仅包含 eng_prefixes 开头的语料
    return [pair for pair in pairs if filterPair(pair)]

def prepareData(lang1, lang2, reverse=False):
    # 读取文件中的数据
    input_lang, output_lang, pairs = readLangs(lang1, lang2, reverse)
    print("Read %s sentence pairs" % len(pairs))
    
    # 按条件选取语料
    pairs = filterPairs(pairs[:])
    print("Trimmed to %s sentence pairs" % len(pairs))
    print("Counting words...")
    
    # 将语料保存至相应的语言类
    for pair in pairs:
        input_lang.addSentence(pair[0])
        output_lang.addSentence(pair[1])
        
    # 打印语言类的信息    
    print("Counted words:")
    print(input_lang.name, input_lang.n_words)
    print(output_lang.name, output_lang.n_words)
    return input_lang, output_lang, pairs

input_lang, output_lang, pairs = prepareData('eng', 'fra', True)
print(random.choice(pairs))

代码输出

Reading lines...
Read 135842 sentence pairs
Trimmed to 10599 sentence pairs
Counting words...
Counted words:
fra 4345
eng 2803
['je volerai vers la lune .', 'i m going to fly to the moon .']

二、Seq2Seq 模型

1. 编码器（Encoder）

class EncoderRNN(nn.Module):
    def __init__(self, input_size, hidden_size):
        super(EncoderRNN, self).__init__()
        self.hidden_size = hidden_size
        self.embedding   = nn.Embedding(input_size, hidden_size)
        self.gru         = nn.GRU(hidden_size, hidden_size)

    def forward(self, input, hidden):
        embedded       = self.embedding(input).view(1, 1, -1)
        output         = embedded
        output, hidden = self.gru(output, hidden)
        return output, hidden

    def initHidden(self):
        return torch.zeros(1, 1, self.hidden_size, device=device)

2. 解码器（Decoder）

class AttnDecoderRNN(nn.Module):
    def __init__(self, hidden_size, output_size, dropout_p=0.1, max_length=MAX_LENGTH):
        super(AttnDecoderRNN, self).__init__()
        self.hidden_size = hidden_size
        self.output_size = output_size
        self.dropout_p = dropout_p
        self.max_length = max_length

        self.embedding = nn.Embedding(self.output_size, self.hidden_size)
        self.attn = nn.Linear(self.hidden_size * 2, self.max_length)
        self.attn_combine = nn.Linear(self.hidden_size * 2, self.hidden_size)
        self.dropout = nn.Dropout(self.dropout_p)
        self.gru = nn.GRU(self.hidden_size, self.hidden_size)
        self.out = nn.Linear(self.hidden_size, self.output_size)

    def forward(self, input, hidden, encoder_outputs):
        embedded = self.embedding(input).view(1, 1, -1)
        embedded = self.dropout(embedded)

        attn_weights = F.softmax(
            self.attn(torch.cat((embedded[0], hidden[0]), 1)), dim=1)
        attn_applied = torch.bmm(attn_weights.unsqueeze(0),
                                 encoder_outputs.unsqueeze(0))

        output = torch.cat((embedded[0], attn_applied[0]), 1)
        output = self.attn_combine(output).unsqueeze(0)

        output = F.relu(output)
        output, hidden = self.gru(output, hidden)

        output = F.log_softmax(self.out(output[0]), dim=1)
        return output, hidden, attn_weights

    def initHidden(self):
        return torch.zeros(1, 1, self.hidden_size, device=device)

三、训练

1. 数据预处理

# 将文本数字化，获取词汇index
def indexesFromSentence(lang, sentence):
    return [lang.word2index[word] for word in sentence.split(' ')]

# 将数字化的文本，转化为tensor数据
def tensorFromSentence(lang, sentence):
    indexes = indexesFromSentence(lang, sentence)
    indexes.append(EOS_token)
    return torch.tensor(indexes, dtype=torch.long, device=device).view(-1, 1)

# 输入pair文本，输出预处理好的数据
def tensorsFromPair(pair):
    input_tensor  = tensorFromSentence(input_lang, pair[0])
    target_tensor = tensorFromSentence(output_lang, pair[1])
    return (input_tensor, target_tensor)

2. 训练函数

teacher_forcing_ratio = 0.5

def train(input_tensor, target_tensor, 
          encoder, decoder, 
          encoder_optimizer, decoder_optimizer, 
          criterion, max_length=MAX_LENGTH):
    
    # 编码器初始化
    encoder_hidden = encoder.initHidden()
    
    # grad属性归零
    encoder_optimizer.zero_grad()
    decoder_optimizer.zero_grad()

    input_length  = input_tensor.size(0)
    target_length = target_tensor.size(0)
    
    # 用于创建一个指定大小的全零张量（tensor），用作默认编码器输出
    encoder_outputs = torch.zeros(max_length, encoder.hidden_size, device=device)

    loss = 0
    
    # 将处理好的语料送入编码器
    for ei in range(input_length):
        encoder_output, encoder_hidden = encoder(input_tensor[ei], encoder_hidden)
        encoder_outputs[ei]            = encoder_output[0, 0]
    
    # 解码器默认输出
    decoder_input  = torch.tensor([[SOS_token]], device=device)
    decoder_hidden = encoder_hidden

    use_teacher_forcing = True if random.random() < teacher_forcing_ratio else False
    
    # 将编码器处理好的输出送入解码器
    if use_teacher_forcing:
        # Teacher forcing: Feed the target as the next input
        for di in range(target_length):
            decoder_output, decoder_hidden, decoder_attention = decoder(
                decoder_input, decoder_hidden, encoder_outputs)
            
            loss         += criterion(decoder_output, target_tensor[di])
            decoder_input = target_tensor[di]  # Teacher forcing
    else:
        # Without teacher forcing: use its own predictions as the next input
        for di in range(target_length):
            decoder_output, decoder_hidden, decoder_attention = decoder(
                decoder_input, decoder_hidden, encoder_outputs)
            
            topv, topi    = decoder_output.topk(1)
            decoder_input = topi.squeeze().detach()  # detach from history as input

            loss         += criterion(decoder_output, target_tensor[di])
            if decoder_input.item() == EOS_token:
                break

    loss.backward()

    encoder_optimizer.step()
    decoder_optimizer.step()

    return loss.item() / target_length

在序列生成的任务中，如机器翻译或文本生成，解码器（decoder）的输入通常是由解码器自己生成的预测结果，即前一个时间步的输出。然而，这种自回归方式可能存在一个问题，即在训练过程中，解码器可能会产生累积误差，并导致输出与目标序列逐渐偏离。
为了解决这个问题，引入了一种称为"Teacher Forcing"的技术。在训练过程中，Teacher Forcing将目标序列的真实值作为解码器的输入，而不是使用解码器自己的预测结果。这样可以提供更准确的指导信号，帮助解码器更快地学习到正确的输出。
在这段代码中，use_teacher_forcing变量用于确定解码器在训练阶段使用何种策略作为下一个输入。
当use_teacher_forcing为True时，采用"Teacher Forcing"的策略，即将目标序列中的真实标签作为解码器的下一个输入。而当use_teacher_forcing为False时，采用"Without Teacher Forcing"的策略，即将解码器自身的预测作为下一个输入。
使用use_teacher_forcing的目的是在训练过程中平衡解码器的预测能力和稳定性。以下是对两种策略的解释：

1. Teacher Forcing: 在每个时间步（di循环中），解码器的输入都是目标序列中的真实标签。这样做的好处是，解码器可以直接获得正确的输入信息，加快训练速度，并且在训练早期提供更准确的梯度信号，帮助解码器更好地学习。然而，过度依赖目标序列可能会导致模型过于敏感，一旦目标序列中出现错误，可能会在解码器中产生累积的误差。
2. Without Teacher Forcing: 在每个时间步，解码器的输入是前一个时间步的预测输出。这样做的好处是，解码器需要依靠自身的预测能力来生成下一个输入，从而更好地适应真实应用场景中可能出现的输入变化。这种策略可以提高模型的稳定性，但可能会导致训练过程更加困难，特别是在初始阶段。

一般来说，Teacher Forcing策略在训练过程中可以帮助模型快速收敛，而Without Teacher Forcing策略则更接近真实应用中的生成场景。通常会使用一定比例的Teacher Forcing，在训练过程中逐渐减小这个比例，以便模型逐渐过渡到更自主的生成模式。
综上所述，通过使用use_teacher_forcing来选择不同的策略，可以在训练解码器时平衡模型的预测能力和稳定性，同时也提供了更灵活的生成模式选择。

1. topv, topi = decoder_output.topk(1)
   这一行代码使用.topk(1)函数从decoder_output中获取最大的元素及其对应的索引。decoder_output是一个张量（tensor），它包含了解码器的输出结果，可能是一个概率分布或是其他的数值。.topk(1)函数将返回两个张量：topv和topi。topv是最大的元素值，而topi是对应的索引值。
2. decoder_input = topi.squeeze().detach() 这一行代码对topi进行处理，以便作为下一个解码器的输入。首先，.squeeze()函数被调用，它的作用是去除张量中维度为1的维度，从而将topi的形状进行压缩。然后，.detach()函数被调用，它的作用是将张量从计算图中分离出来，使得在后续的计算中不会对该张量进行梯度计算。最后，将处理后的张量赋值给decoder_input，作为下一个解码器的输入。

import time
import math

def asMinutes(s):
    m = math.floor(s / 60)
    s -= m * 60
    return '%dm %ds' % (m, s)

def timeSince(since, percent):
    now = time.time()
    s = now - since
    es = s / (percent)
    rs = es - s
    return '%s (- %s)' % (asMinutes(s), asMinutes(rs))

def trainIters(encoder,decoder,n_iters,print_every=1000,
               plot_every=100,learning_rate=0.01):
    
    start = time.time()
    plot_losses      = []
    print_loss_total = 0  # Reset every print_every
    plot_loss_total  = 0  # Reset every plot_every

    encoder_optimizer = optim.SGD(encoder.parameters(), lr=learning_rate)
    decoder_optimizer = optim.SGD(decoder.parameters(), lr=learning_rate)
    
    # 在 pairs 中随机选取 n_iters 条数据用作训练集
    training_pairs    = [tensorsFromPair(random.choice(pairs)) for i in range(n_iters)]
    criterion         = nn.NLLLoss()

    for iter in range(1, n_iters + 1):
        training_pair = training_pairs[iter - 1]
        input_tensor  = training_pair[0]
        target_tensor = training_pair[1]

        loss = train(input_tensor, target_tensor, encoder,
                     decoder, encoder_optimizer, decoder_optimizer, criterion)
        print_loss_total += loss
        plot_loss_total  += loss

        if iter % print_every == 0:
            print_loss_avg   = print_loss_total / print_every
            print_loss_total = 0
            print('%s (%d %d%%) %.4f' % (timeSince(start, iter / n_iters),
                                         iter, iter / n_iters * 100, print_loss_avg))

        if iter % plot_every == 0:
            plot_loss_avg = plot_loss_total / plot_every
            plot_losses.append(plot_loss_avg)
            plot_loss_total = 0

    return plot_losses

3. 评估

def evaluate(encoder, decoder, sentence, max_length=MAX_LENGTH):
    with torch.no_grad():
        input_tensor    = tensorFromSentence(input_lang, sentence)
        input_length    = input_tensor.size()[0]
        encoder_hidden  = encoder.initHidden()

        encoder_outputs = torch.zeros(max_length, encoder.hidden_size, device=device)

        for ei in range(input_length):
            encoder_output, encoder_hidden = encoder(input_tensor[ei],encoder_hidden)
            encoder_outputs[ei]           += encoder_output[0, 0]

        decoder_input  = torch.tensor([[SOS_token]], device=device)  # SOS

        decoder_hidden = encoder_hidden

        decoded_words  = []
        decoder_attentions = torch.zeros(max_length, max_length)

        for di in range(max_length):
            decoder_output, decoder_hidden, decoder_attention = decoder(
                decoder_input, decoder_hidden, encoder_outputs)
            
            decoder_attentions[di] = decoder_attention.data
            topv, topi             = decoder_output.data.topk(1)
            
            if topi.item() == EOS_token:
                decoded_words.append('<EOS>')
                break
            else:
                decoded_words.append(output_lang.index2word[topi.item()])

            decoder_input = topi.squeeze().detach()

        return decoded_words, decoder_attentions[:di + 1]

def evaluateRandomly(encoder, decoder, n=5):
    for i in range(n):
        pair = random.choice(pairs)
        print('>', pair[0])
        print('=', pair[1])
        output_words, attentions = evaluate(encoder, decoder, pair[0])
        output_sentence = ' '.join(output_words)
        print('<', output_sentence)
        print('')

四、训练与评估

hidden_size   = 256
encoder1      = EncoderRNN(input_lang.n_words, hidden_size).to(device)
attn_decoder1 = AttnDecoderRNN(hidden_size, output_lang.n_words, dropout_p=0.1).to(device)

plot_losses   = trainIters(encoder1, attn_decoder1, 10000, print_every=5000)

代码输出

6m 41s (- 6m 41s) (5000 50%) 2.8497
13m 28s (- 0m 0s) (10000 100%) 2.2939

evaluateRandomly(encoder1, attn_decoder1)

代码输出

> tu es en grave danger .
= you re in serious danger .
< you are the of . . <EOS>

> il est parfait pour le poste .
= he is just right for the job .
< he is out to the . . <EOS>

> je te quitte demain .
= i m leaving you tomorrow .
< i am glad to . . <EOS>

> c est un auteur .
= he s an author .
< he s a good . <EOS>

> nous sommes des prisonniers .
= we re prisoners .
< we re in . <EOS>

1. Loss图

import matplotlib.pyplot as plt
#隐藏警告
import warnings
warnings.filterwarnings("ignore")               # 忽略警告信息
# plt.rcParams['font.sans-serif']    = ['SimHei'] # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False      # 用来正常显示负号
plt.rcParams['figure.dpi']         = 100        # 分辨率

epochs_range = range(len(plot_losses))

plt.figure(figsize=(8, 3))

plt.subplot(1, 1, 1)
plt.plot(epochs_range, plot_losses, label='Training Loss')
plt.legend(loc='upper right')
plt.title('Training Loss')
plt.show()

代码输出

2. 可视化注意力

import matplotlib.pyplot as plt

output_words, attentions = evaluate(encoder1, attn_decoder1, "je suis trop froid .")
plt.matshow(attentions.numpy())

代码输出

<matplotlib.image.AxesImage at 0x1f912b9d600>

import matplotlib.ticker as ticker
#隐藏警告
import warnings
warnings.filterwarnings("ignore")               # 忽略警告信息

def showAttention(input_sentence, output_words, attentions):
    # Set up figure with colorbar
    fig = plt.figure()
    ax = fig.add_subplot(111)
    cax = ax.matshow(attentions.numpy(), cmap='bone')
    fig.colorbar(cax)

    # Set up axes
    ax.set_xticklabels([''] + input_sentence.split(' ') +
                       ['<EOS>'], rotation=90)
    ax.set_yticklabels([''] + output_words)

    # Show label at every tick
    ax.xaxis.set_major_locator(ticker.MultipleLocator(1))
    ax.yaxis.set_major_locator(ticker.MultipleLocator(1))

    plt.show()

def evaluateAndShowAttention(input_sentence):
    output_words, attentions = evaluate(
        encoder1, attn_decoder1, input_sentence)
    print('input =', input_sentence)
    print('output =', ' '.join(output_words))
    showAttention(input_sentence, output_words, attentions)


evaluateAndShowAttention("elle a cinq ans de moins que moi .")
evaluateAndShowAttention("elle est trop petit .")
evaluateAndShowAttention("je ne crains pas de mourir .")
evaluateAndShowAttention("c est un jeune directeur plein de talent .")

代码输出（下面的内容全都是代码运行输出的结果）

input = elle a cinq ans de moins que moi .
output = she s taller than me than me me . .

input = elle est trop petit .
output = she s too old . <EOS>

input = je ne crains pas de mourir .
output = i m not going to . . . <EOS>

input = c est un jeune directeur plein de talent .
output = he s a good at . . <EOS>