带注意力机制的seq2seq
理论
回忆一下seq2seq,编码器的输出了一个state给解码器,context = state[-1].repeat(X.shape[0], 1, 1)
,解码器吧state作为上下文对象和解码器输入一起并入丢到RNN中。
seq2seq解码器
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):
return enc_outputs[1]
def forward(self, X, state):
# 输出'X'的形状:(batch_size,num_steps,embed_size)
# permute(1, 0, 2)把'num_steps'和'num_steps'换一下
X = self.embedding(X).permute(1, 0, 2)
# 广播context,使其具有与X相同的num_steps
context = state[-1].repeat(X.shape[0], 1, 1)
X_and_context = torch.cat((X, context), 2)
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[0]的形状:(num_layers,batch_size,num_hiddens)
return output, state
这样有个问题:
编码器的state真的是解码器最需要的信息吗?对于机器翻译这种任务,对应词的信息是比较关键的,比如英->法,翻译bonjour的时候,解码器更想看到编码器中‘hello’经过RNN输出的隐藏状态,这个问题,用attention就可以解决。
k,v:编码器RNN的输出
q:解码器上一个时间步的输出隐藏状态,和k,v计算Attention再进入RNN得到输出
代码
import torch
from torch import nn
from d2l import torch as d2l
#@save
class AttentionDecoder(d2l.Decoder):
"""带有注意力机制解码器的基本接口"""
def __init__(self, **kwargs):
super(AttentionDecoder, self).__init__(**kwargs)
@property
def attention_weights(self):
raise NotImplementedError
编码器区别不大,主要区别在解码器
class Seq2SeqAttentionDecoder(AttentionDecoder):
def __init__(self, vocab_size, embed_size, num_hiddens, num_layers,
dropout=0, **kwargs):
super(Seq2SeqAttentionDecoder, self).__init__(**kwargs)
self.attention = d2l.AdditiveAttention(
num_hiddens, num_hiddens, num_hiddens, dropout)
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, enc_valid_lens, *args):
# outputs的形状为(batch_size,num_steps,num_hiddens).
# hidden_state的形状为(num_layers,batch_size,num_hiddens)
outputs, hidden_state = enc_outputs
return (outputs.permute(1, 0, 2), hidden_state, enc_valid_lens)
def forward(self, X, state):
# enc_outputs的形状为(batch_size,num_steps,num_hiddens).
# hidden_state的形状为(num_layers,batch_size,num_hiddens)
enc_outputs, hidden_state, enc_valid_lens = state
# 输出X的形状为(num_steps,batch_size,embed_size)
X = self.embedding(X).permute(1, 0, 2)
outputs, self._attention_weights = [], []
for x in X:
# query的形状为(batch_size,1,num_hiddens)
query = torch.unsqueeze(hidden_state[-1], dim=1)
# context的形状为(batch_size,1,num_hiddens)
context = self.attention(query, enc_outputs, enc_outputs, enc_valid_lens)
# 在特征维度上连结
x = torch.cat((context, torch.unsqueeze(x, dim=1)), dim=-1)
# 将x变形为(1,batch_size,embed_size+num_hiddens)
out, hidden_state = self.rnn(x.permute(1, 0, 2), hidden_state)
outputs.append(out)
self._attention_weights.append(self.attention.attention_weights)
# 全连接层变换后,outputs的形状为
# (num_steps,batch_size,vocab_size)
outputs = self.dense(torch.cat(outputs, dim=0))
return outputs.permute(1, 0, 2), [enc_outputs, hidden_state, enc_valid_lens]
@property
def attention_weights(self):
return self._attention_weights
对于解码器的每个
批量
X输入,都对他做Attention
query = torch.unsqueeze(hidden_state[-1], dim=1)
hidden_state[-1](批量大小,隐藏单元数)是编码器的输出state拿过来,因为编码器是两层的RNN所以取[-1] ,然后增加一维
上下文的构造
context = self.attention(query, enc_outputs, enc_outputs, enc_valid_lens)
key和value是enc_outputs编码器输出(时间步数,批量大小,隐藏单元数)编码器的输出在传入解码器使用init_state方法时,做了维度转换
outputs.permute(1, 0, 2)
(批量大小,时间步数,隐藏单元数) 诶,这就对上了~之后上下文拼接上解码器输入x,剩下的就很简单能看懂了
总结一下:
query:编码器输出的隐藏状态(批量大小,特征长度)->(批量大小,1,特征长度)
特征长度
在Attention中和W矩阵做乘法就被映射成隐藏单元数
了(批量大小,1,隐藏单元个数)key、value:编码器的输出(时间步数,批量大小,隐藏单元数)->(批量大小,时间步数,隐藏单元数)
key在Attention中被映射成(批量大小,,隐藏单元数)
value没变
context:做完Attention之后的值(批量大小,query的时间步1,value的特征长度)
X:解码器自己的输入(批量大小,时间步,特征长度)->(时间步,批量大小,特征长度)
x:每个时间步的输入(批量大小,特征长度)
丢入解码器RNN得到下一个query
outputs = self.dense(torch.cat(outputs, dim=0))
最后的输出是把所有的输出拼在一起丢到线性层输出,why?
测试一下
encoder = d2l.Seq2SeqEncoder(vocab_size=10, embed_size=8, num_hiddens=16,
num_layers=2)
encoder.eval()
decoder = Seq2SeqAttentionDecoder(vocab_size=10, embed_size=8, num_hiddens=16,
num_layers=2)
decoder.eval()
X = torch.zeros((4, 7), dtype=torch.long) # (batch_size,num_steps)
state = decoder.init_state(encoder(X), None)
output, state = decoder(X, state)
output.shape, len(state), state[0].shape, len(state[1]), state[1][0].shape
输出
(torch.Size([4, 7, 10]), 3, torch.Size([4, 7, 16]), 2, torch.Size([4, 16]))
训练
embed_size, num_hiddens, num_layers, dropout = 32, 32, 2, 0.1
batch_size, num_steps = 64, 10
lr, num_epochs, device = 0.005, 250, d2l.try_gpu()
train_iter, src_vocab, tgt_vocab = d2l.load_data_nmt(batch_size, num_steps)
encoder = d2l.Seq2SeqEncoder(
len(src_vocab), embed_size, num_hiddens, num_layers, dropout)
decoder = Seq2SeqAttentionDecoder(
len(tgt_vocab), embed_size, num_hiddens, num_layers, dropout)
net = d2l.EncoderDecoder(encoder, decoder)
d2l.train_seq2seq(net, train_iter, lr, num_epochs, tgt_vocab, device)
loss 0.020, 4423.3 tokens/sec on cuda:0
测试翻译
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, dec_attention_weight_seq = d2l.predict_seq2seq(
net, eng, src_vocab, tgt_vocab, num_steps, device, True)
print(f'{eng} => {translation}, ',
f'bleu {d2l.bleu(translation, fra, k=2):.3f}')
go . => va !, bleu 1.000
i lost . => j'ai perdu ., bleu 1.000
he's calm . => je suis chez moi ., bleu 0.000
i'm home . => je suis chez moi ., bleu 1.000
画一下权重的图
attention_weights = torch.cat([step[0][0][0] for step in dec_attention_weight_seq], 0).reshape((
1, 1, -1, num_steps))
# 加上一个包含序列结束词元
d2l.show_heatmaps(
attention_weights[:, :, :, :len(engs[-1].split()) + 1].cpu(),
xlabel='Key positions', ylabel='Query positions')