(一)序列To序列 / Seq2Seq
这是什么意思呢?也就是说输入是一串序列,输出也是一串序列。就像是生物学中的DNA转录成RNA序列一样。其实他最早被用于机器翻译,给定一个源语言的句子,自动翻译成目标语言。两个序列可以有不同的长度。
序列To序列 / Seq2Seq,使用的也是编码器,解码器架构。编码器是一个RNN,读取输入句子,解码器使用另一个RNN来输出。并且双向循环神经网络可以用于编码器。
(1)编码器解码器细节
上一篇介绍了什么是编码器-解码器架构,这里我们仔细观察一下两者之间的连接。
- 编码器是没有输出的RNN,编码器的最后一层会将隐状态输送到解码器的RNN作为输入,并且解码器有自己的输入
- 编码器和解码器的RNN是两个不同的网络
- 编码器最后的时间步隐状态用作解码器的初始隐状态
(2)训练和推理(预测)
训练时解码器使用目标句子作为输入。
推理的时候,它是由上一个时刻的输出作为下一个时刻的输入,不断迭代下去。
具体实现见代码。
(3)判断预测句子的好坏
衡量序列的好坏——BLEU。
是预测中所有n-gram的精度,n-gram就是n元的意思,就是在序列中连续的子序列的个数。就是说不论顺序,预测值的n-gram的组合在真实n-gram中的比例。
例如:
标签序列:A,B,C,D,E,F
预测序列:A,B,B,C,D,E
则:、、、
BLEU的定义:(越大越好)
(二)代码实现
import collections
import math
import torch
from torch import nn
from d2l import torch as d2l
注意这里的输入的形状是(batch_size,num_steps),每个num_step是用下标表示的一个词。
然后embedding的参数列表(词的总个数,编码维度)用来实现词与词向量的映射。
例如:nn.Embedding(28, 2),一共有28个词,每个词用长度为2的向量表示。
给一个(10,10)的输入,批量为10,时间步设为10的输入,输出就会是(10,10,2)他将里面的所有词都转变为二维向量。
embedding = nn.Embedding(10, 2) # 10个词,每个词用2维词向量表示
input = torch.arange(0, 6).view(3, 2).long() # 3个句子,每句子有2个词
# input = t.autograd.Variable(input)
output = embedding(input)
print(output.size())
print(embedding.weight.size())
# 编码器
#@save
class Seq2SeqEncoder(d2l.Encoder):
"""用于序列到序列学习的循环神经网络编码器"""
def __init__(self, vocab_size, embed_size, num_hiddens, num_layers,dropout=0, **kwargs):
super(Seq2SeqEncoder, self).__init__(**kwargs)
# 嵌入层
self.embedding = nn.Embedding(vocab_size, embed_size)
self.rnn = nn.GRU(embed_size, num_hiddens, num_layers,
dropout=dropout)
def forward(self, X, *args):
# 输入'X'的形状:(batch_size,num_steps)
# 输出'X'的形状:(batch_size,num_steps,embed_size)
X = self.embedding(X)
# 在循环神经网络模型中,第一个轴对应于时间步
X = X.permute(1, 0, 2)
# 如果未提及状态,则默认为0
output, state = self.rnn(X)
# output的形状:(num_steps,batch_size,num_hiddens)
# state的形状:(num_layers,batch_size,num_hiddens)
return output, state
encoder = Seq2SeqEncoder(vocab_size=10, embed_size=8, num_hiddens=16,
num_layers=2)
encoder.eval()
X = torch.zeros((4, 7), dtype=torch.long)
output, state = encoder(X)
output.shape, state.shape
class Seq2SeqDecoder(d2l.Decoder):
"""用于序列到序列学习的循环神经网络解码器"""
def __init__(self, vocab_size, embed_size, num_hiddens, num_layers,
dropout=0, **kwargs):
super(Seq2SeqDecoder, self).__init__(**kwargs)
self.embedding = nn.Embedding(vocab_size, embed_size)
self.rnn = nn.GRU(embed_size + num_hiddens, num_hiddens, num_layers,dropout=dropout)
self.dense = nn.Linear(num_hiddens, vocab_size)
def init_state(self, enc_outputs, *args):
# outputs是指(output, state)
return enc_outputs[1]
def forward(self, X, state):
# 输入'X'的形状:(batch_size,num_steps)
# 输出'X'的形状:(batch_size,num_steps,embed_size)
X = self.embedding(X).permute(1, 0, 2) # (num_steps,batch_size,embed_size)
# 广播context,使其具有与X相同的num_steps (num_steps,batch_size,num_hiddens)
context = state[-1].repeat(X.shape[0], 1, 1) # (通道数的重复倍数,列的重复倍数,行的重复倍数)
# context的形状(num_steps,batch_size,num_hiddens)
X_and_context = torch.cat((X, context), 2) # 这就是为什么GRU的输入长度是embed_size + num_hiddens
output, state = self.rnn(X_and_context, state)
output = self.dense(output).permute(1, 0, 2)
# output的形状:(batch_size,num_steps,vocab_size)
# state的形状:(num_layers,batch_size,num_hiddens)
return output, state
decoder = Seq2SeqDecoder(vocab_size=10, embed_size=8, num_hiddens=16,
num_layers=2)
decoder.eval()
state = decoder.init_state(encoder(X))
# print(state.shape) # torch.Size([2, 4, 16])
output, state = decoder(X, state)
output.shape, state.shape, X.shape
# 通过零值化,屏蔽不相关项。
def sequence_mask(X, valid_len, value=0):
"""在序列中屏蔽不相关的项"""
maxlen = X.size(1)
# print(maxlen) # 3
# [None,:]是按行增加一个维度,[:, None]按列增加一个维度
mask = torch.arange((maxlen), dtype=torch.float32,device=X.device)[None, :] < valid_len[:, None]
# print(torch.arange((maxlen), dtype=torch.float32,device=X.device)[None, :]) # tensor([[0., 1., 2.]])
# print(valid_len[:, None])
# tensor([[1],
# [2]])
X[~mask] = value
return X
X = torch.tensor([[1, 2, 3], [4, 5, 6]])
sequence_mask(X, torch.tensor([1, 2]))
class MaskedSoftmaxCELoss(nn.CrossEntropyLoss):
"""带遮蔽的softmax交叉熵损失函数"""
# pred的形状:(batch_size,num_steps,vocab_size)
# label的形状:(batch_size,num_steps)
# valid_len的形状:(batch_size,)
def forward(self, pred, label, valid_len):
weights = torch.ones_like(label)
weights = sequence_mask(weights, valid_len)# 只留下有效长度内权重,其他的置为0
self.reduction='none'
unweighted_loss = super(MaskedSoftmaxCELoss, self).forward(pred.permute(0, 2, 1), label) # permute后的形状(batch_size,vocab_size,num_steps)
weighted_loss = (unweighted_loss * weights).mean(dim=1)
return weighted_loss
loss = MaskedSoftmaxCELoss()
print(loss(torch.ones((3, 4, 10), dtype=torch.float32), torch.ones((3, 4), dtype=torch.long), torch.tensor([4, 2, 0])))
l = nn.CrossEntropyLoss(reduction="none")
print(l(torch.ones((3, 10, 4), dtype=torch.float32), torch.ones((3, 4), dtype=torch.long)))
def train_seq2seq(net, data_iter, lr, num_epochs, tgt_vocab, device):
"""训练序列到序列模型"""
def xavier_init_weights(m):
if type(m) == nn.Linear:
nn.init.xavier_uniform_(m.weight)
if type(m) == nn.GRU:
for param in m._flat_weights_names:
if "weight" in param:
nn.init.xavier_uniform_(m._parameters[param])
net.apply(xavier_init_weights)
net.to(device)
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
loss = MaskedSoftmaxCELoss()
net.train()
animator = d2l.Animator(xlabel='epoch', ylabel='loss',
xlim=[10, num_epochs])
for epoch in range(num_epochs):
timer = d2l.Timer()
metric = d2l.Accumulator(2) # 训练损失总和,词元数量
for batch in data_iter:
optimizer.zero_grad()
X, X_valid_len, Y, Y_valid_len = [x.to(device) for x in batch]
bos = torch.tensor([tgt_vocab['<bos>']] * Y.shape[0],
device=device).reshape(-1, 1)# (batch_size, 1)
dec_input = torch.cat([bos, Y[:, :-1]], 1) # 强制教学
# print(bos.shape,dec_input.shape) # torch.Size([64, 1]) torch.Size([64, 10])
Y_hat, _ = net(X, dec_input, X_valid_len)
l = loss(Y_hat, Y, Y_valid_len)
l.sum().backward() # 损失函数的标量进行“反向传播”
d2l.grad_clipping(net, 1)
num_tokens = Y_valid_len.sum()
optimizer.step()
with torch.no_grad():
metric.add(l.sum(), num_tokens)
if (epoch + 1) % 10 == 0:
animator.add(epoch + 1, (metric[0] / metric[1],))
print(f'loss {metric[0] / metric[1]:.3f}, {metric[1] / timer.stop():.1f} '
f'tokens/sec on {str(device)}')
class EncoderDecoder(nn.Module):
"""The base class for the encoder-decoder architecture.
Defined in :numref:`sec_encoder-decoder`"""
def __init__(self, encoder, decoder, **kwargs):
super(EncoderDecoder, self).__init__(**kwargs)
self.encoder = encoder
self.decoder = decoder
def forward(self, enc_X, dec_X, *args):
enc_outputs = self.encoder(enc_X, *args)
dec_state = self.decoder.init_state(enc_outputs, *args)
return self.decoder(dec_X, dec_state)
embed_size, num_hiddens, num_layers, dropout = 32, 32, 2, 0.1
batch_size, num_steps = 64, 10
lr, num_epochs, device = 0.005, 300, d2l.try_gpu()
train_iter, src_vocab, tgt_vocab = d2l.load_data_nmt(batch_size, num_steps)
encoder = Seq2SeqEncoder(len(src_vocab), embed_size, num_hiddens, num_layers,
dropout)
decoder = Seq2SeqDecoder(len(tgt_vocab), embed_size, num_hiddens, num_layers,
dropout)
net = d2l.EncoderDecoder(encoder, decoder)
train_seq2seq(net, train_iter, lr, num_epochs, tgt_vocab, device)
#@save
def predict_seq2seq(net, src_sentence, src_vocab, tgt_vocab, num_steps,
device, save_attention_weights=False):
"""序列到序列模型的预测"""
# 在预测时将net设置为评估模式
net.eval()
src_tokens = src_vocab[src_sentence.lower().split(' ')] + [src_vocab['<eos>']]# 在末尾加上结束标记
enc_valid_len = torch.tensor([len(src_tokens)], device=device)
src_tokens = d2l.truncate_pad(src_tokens, num_steps, src_vocab['<pad>'])# 填充字符
# 添加批量轴
enc_X = torch.unsqueeze(torch.tensor(src_tokens, dtype=torch.long, device=device), dim=0) # 增加一个维度
enc_outputs = net.encoder(enc_X, enc_valid_len)
dec_state = net.decoder.init_state(enc_outputs, enc_valid_len) # 初始化解码器
# 添加批量轴
dec_X = torch.unsqueeze(torch.tensor([tgt_vocab['<bos>']], dtype=torch.long, device=device), dim=0) # 解码器的输入我们放入一个开始标记符
output_seq, attention_weight_seq = [], []
for _ in range(num_steps):
Y, dec_state = net.decoder(dec_X, dec_state)
# 我们使用具有预测最高可能性的词元,作为解码器在下一时间步的输入
# Y的形状:(batch_size,num_steps,vocab_size), batch_size = 1
dec_X = Y.argmax(dim=2) # 每个批量,获取每个num_step最大下标,并作为下一个预测值的开始(batch_size,num_step),下标和tgt_vocab一致
# print(Y.shape) # torch.Size([1, 1, 201])
# print(dec_X.shape) # torch.Size([1, 1])
pred = dec_X.squeeze(dim=0).type(torch.int32).item()# 去掉一个维度
# 保存注意力权重(稍后讨论)
if save_attention_weights:
attention_weight_seq.append(net.decoder.attention_weights)
# 一旦序列结束词元被预测,输出序列的生成就完成了
if pred == tgt_vocab['<eos>']:
break
output_seq.append(pred)
return ' '.join(tgt_vocab.to_tokens(output_seq)), attention_weight_seq
def bleu(pred_seq, label_seq, k): #@save
"""计算BLEU"""
pred_tokens, label_tokens = pred_seq.split(' '), label_seq.split(' ')
len_pred, len_label = len(pred_tokens), len(label_tokens)
score = math.exp(min(0, 1 - len_label / len_pred))
for n in range(1, k + 1):
num_matches, label_subs = 0, collections.defaultdict(int)
for i in range(len_label - n + 1):
label_subs[' '.join(label_tokens[i: i + n])] += 1
for i in range(len_pred - n + 1):
if label_subs[' '.join(pred_tokens[i: i + n])] > 0:
num_matches += 1
label_subs[' '.join(pred_tokens[i: i + n])] -= 1
score *= math.pow(num_matches / (len_pred - n + 1), math.pow(0.5, n))
return score
engs = ['go .', "i lost .", 'he\'s calm .', 'i\'m home .']
fras = ['va !', 'j\'ai perdu .', 'il est calme .', 'je suis chez moi .']
for eng, fra in zip(engs, fras):
translation, attention_weight_seq = predict_seq2seq(
net, eng, src_vocab, tgt_vocab, num_steps, device)
print(f'{eng} => {translation}, bleu {bleu(translation, fra, k=2):.3f}')
网友评论