.. _sec_seq2seq_attention:
Bahdanau 注意力
===============
:numref:`sec_seq2seq`\ 中探讨了机器翻译问题：
通过设计一个基于两个循环神经网络的编码器-解码器架构，
用于序列到序列学习。
具体来说，循环神经网络编码器将长度可变的序列转换为固定形状的上下文变量，
然后循环神经网络解码器根据生成的词元和上下文变量
按词元生成输出（目标）序列词元。
然而，即使并非所有输入（源）词元都对解码某个词元都有用，
在每个解码步骤中仍使用编码\ *相同*\ 的上下文变量。
有什么方法能改变上下文变量呢？
我们试着从 :cite:`Graves.2013`\ 中找到灵感：
在为给定文本序列生成手写的挑战中， Graves设计了一种可微注意力模型，
将文本字符与更长的笔迹对齐， 其中对齐方式仅向一个方向移动。
受学习对齐想法的启发，Bahdanau等人提出了一个没有严格单向对齐限制的
可微注意力模型 :cite:`Bahdanau.Cho.Bengio.2014`\ 。
在预测词元时，如果不是所有输入词元都相关，模型将仅对齐（或参与）输入序列中与当前预测相关的部分。这是通过将上下文变量视为注意力集中的输出来实现的。
模型
----
下面描述的Bahdanau注意力模型 将遵循
:numref:`sec_seq2seq`\ 中的相同符号表达。 这个新的基于注意力的模型与
:numref:`sec_seq2seq`\ 中的模型相同， 只不过
:eq:`eq_seq2seq_s_t`\ 中的上下文变量\ :math:`\mathbf{c}`
在任何解码时间步\ :math:`t'`\ 都会被\ :math:`\mathbf{c}_{t'}`\ 替换。
假设输入序列中有\ :math:`T`\ 个词元，
解码时间步\ :math:`t'`\ 的上下文变量是注意力集中的输出：
.. math:: \mathbf{c}_{t'} = \sum_{t=1}^T \alpha(\mathbf{s}_{t' - 1}, \mathbf{h}_t) \mathbf{h}_t,
其中，时间步\ :math:`t' - 1`\ 时的解码器隐状态\ :math:`\mathbf{s}_{t' - 1}`\ 是查询，
编码器隐状态\ :math:`\mathbf{h}_t`\ 既是键，也是值，
注意力权重\ :math:`\alpha`\ 是使用 :eq:`eq_attn-scoring-alpha`
所定义的加性注意力打分函数计算的。
与
:numref:`fig_seq2seq_details`\ 中的循环神经网络编码器-解码器架构略有不同，
:numref:`fig_s2s_attention_details`\ 描述了Bahdanau注意力的架构。
.. _fig_s2s_attention_details:
.. figure:: ../img/seq2seq-attention-details.svg
一个带有Bahdanau注意力的循环神经网络编码器-解码器模型
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
import mindspore
from mindspore import nn
from d2l import mindspore as d2l
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
import torch
from torch import nn
from d2l import torch as d2l
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
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 = d2l.Embedding(vocab_size, embed_size)
self.rnn = nn.GRU(
embed_size + num_hiddens, num_hiddens, num_layers,
dropout=dropout)
self.dense = d2l.Dense(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.transpose(1, 0, 2), hidden_state, enc_valid_lens)
def construct(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).transpose(1, 0, 2)
outputs, self._attention_weights = [], []
for i in range(X.shape[0]):
x = X[i]
# query的形状为(batch_size,1,num_hiddens)
query = d2l.expand_dims(hidden_state[-1], axis=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 = d2l.concat((context, d2l.expand_dims(x, axis=1)), axis=-1)
# 将x变形为(1,batch_size,embed_size+num_hiddens)
out, hidden_state = self.rnn(x.transpose(1, 0, 2), hidden_state)
outputs.append(out)
self._attention_weights.append(self.attention.attention_weights)
outputs = self.dense(d2l.concat(outputs, axis=0))
return outputs.transpose(1, 0, 2), (enc_outputs, hidden_state,
enc_valid_lens)
@property
def attention_weights(self):
return self._attention_weights
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
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
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
encoder = d2l.Seq2SeqEncoder(vocab_size=10, embed_size=8, num_hiddens=16,
num_layers=2)
encoder.set_train(False)
decoder = Seq2SeqAttentionDecoder(vocab_size=10, embed_size=8, num_hiddens=16,
num_layers=2)
decoder.set_train(False)
X = d2l.zeros((4, 7), dtype=mindspore.int32)
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
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
((4, 7, 10), 3, (4, 7, 16), 2, (4, 16))
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
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
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
(torch.Size([4, 7, 10]), 3, torch.Size([4, 7, 16]), 2, torch.Size([4, 16]))
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
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)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
loss 0.009, 4896.2 tokens/sec
.. figure:: output_bahdanau-attention_7f08d9_32_1.png
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
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)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
loss 0.020, 24205.7 tokens/sec on cuda:0
.. figure:: output_bahdanau-attention_7f08d9_35_1.svg
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
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, save_attention_weights=True)
print(f'{eng} => {translation}, ',
f'bleu {d2l.bleu(translation, fra, k=2):.3f}')
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
go . => va !, bleu 1.000
i lost . => j'ai perdu ., bleu 1.000
he's calm . => il est mouillé ., bleu 0.658
i'm home . => je suis chez moi ., bleu 1.000
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
attention_weights = d2l.concat([step[0][0][0] for step in dec_attention_weight_seq], 0).reshape((
1, 1, -1, num_steps))
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
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}')
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
go . => va !, bleu 1.000
i lost . => j'ai perdu ., bleu 1.000
he's calm . => j'en maintenant ., bleu 0.000
i'm home . => je suis chez moi ., bleu 1.000
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
attention_weights = torch.cat([step[0][0][0] for step in dec_attention_weight_seq], 0).reshape((
1, 1, -1, num_steps))
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
# 加上一个包含序列结束词元
d2l.show_heatmaps(
attention_weights[:, :, :, :len(engs[-1].split()) + 1],
xlabel='Key positions', ylabel='Query positions')
.. figure:: output_bahdanau-attention_7f08d9_52_0.png
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
# 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')
.. figure:: output_bahdanau-attention_7f08d9_55_0.svg
.. raw:: html
.. raw:: html
.. raw:: html
`讨论 `__
.. raw:: html
.. raw:: html
`讨论 `__
.. raw:: html
.. raw:: html