Deep Learning Note 37 注意力评分(Attention Score)

import math
import torch
from torch import nn
from d2l import torch as d2l

# 遮掩softmax操作
def masked_softmax(X, valid_lens):
    """通过最后一个轴上遮蔽元素来执行 softmax 操作"""
    # X: 3D张量，valid_lens: 1D或2D张量
    if valid_lens is None:
        return nn.functional.softmax(X, dim=-1)
    else:
        shape = X.shape
        if valid_lens.dim() == 1:
            valid_lens = torch.repeat_interleave(valid_lens, shape[1])  # 各元素重复shape[1]次
        else:
            valid_lens = valid_lens.reshape(-1)  # 将valid_lens重塑为1D张量
        # 最后一个轴上被遮蔽的元素使用一个非常大的负值替换，从而其softmax输出为0
        X = d2l.sequence_mask(X.reshape(-1, shape[-1]), valid_lens,
                              value=-1e6)
        return nn.functional.softmax(X.reshape(shape), dim=-1)

# 加性注意力
class AdditiveAttention(nn.Module):
    """加性注意力"""

    def __init__(self, key_size, query_size, num_hiddens, dropout, **kwargs):
        super(AdditiveAttention, self).__init__(**kwargs)
        self.W_k = nn.Linear(key_size, num_hiddens, bias=False)
        self.W_q = nn.Linear(query_size, num_hiddens, bias=False)
        self.w_v = nn.Linear(num_hiddens, 1, bias=False)
        self.dropout = nn.Dropout(dropout)

    def forward(self, queries, keys, values, valid_lens):
        queries, keys = self.W_q(queries), self.W_k(keys)
        # 在维度扩展后
        # queries的形状:(batch_size，查询数量，1，num_hidden)
        # key的形状:(batch_size，1，“键-值”对的数量，num_hiddens)
        # 使用广播方式进行求和
        features = queries.unsqueeze(2) + keys.unsqueeze(1)
        features = torch.tanh(features)
        # self.w_v只有一个输出，因此从形状中移除最后那个维度
        # scores的形状为(batch_size，查询数量，“键-值”对的数量)
        scores = self.w_v(features).squeeze(-1)
        self.attention_weights = masked_softmax(scores, valid_lens)
        # values的形状为(batch_size，“键-值”对的数量，值的维度)
        return torch.bmm(self.dropout(self.attention_weights), values)
        # 最终返回的形状为(batch_size，查询数量，值的维度)
        # 如(batch_num, query_num, value_dim)表示的就是第query_num个query与第value_dim个value的注意力权重

queries, keys = torch.normal(0, 1, (2, 1, 20)), torch.ones((2, 10, 2))
values = torch.arange(40, dtype=torch.float32).reshape(1, 10, 4).repeat(
    2, 1, 1)
valid_lens = torch.tensor([2, 6])
attention = AdditiveAttention(key_size=2, query_size=20, num_hiddens=8,
                              dropout=0.1)
attention.eval()
print(attention(queries, keys, values, valid_lens))

d2l.show_heatmaps(attention.attention_weights.reshape((1, 1, 2, 10)),
                  xlabel='Keys', ylabel='Queries')
d2l.plt.show()

# 缩放点积注意力
class DotProductAttention(nn.Module):
    """缩放点积注意力"""

    def __init__(self, dropout, **kwargs):
        super(DotProductAttention, self).__init__(**kwargs)
        self.dropout = nn.Dropout(dropout)

    # queries的形状：(batch_size，查询数量，d)
    # keys的形状：(batch_size，“键-值”对的数量，d)
    # values的形状：(batch_size，“键-值”对的数量，值的维度)
    # valid_lens的形状：(batch_size，)或(batch_size，查询数量）
    def forward(self, queries, keys, values, valid_lens=None):
        d = queries.shape[-1]
        scores = torch.bmm(queries, keys.transpose(1, 2)) / math.sqrt(d)
        self.attention_weights = masked_softmax(scores, valid_lens)
        return torch.bmm(self.dropout(self.attention_weights), values)