BiLSTM+Attention代码实现（pytorch）

import torch
import torch.nn as nn

class BiLSTM_Attention(nn.Module):
    def __init__(self):
        super(BiLSTM_Attention, self).__init__()

        self.embedding = nn.Embedding(vocab_size, embedding_dim)
        self.lstm = nn.LSTM(embedding_dim, n_hidden, bidirectional=True)
        self.out = nn.Linear(n_hidden * 2, num_classes)

    # lstm_output : [batch_size, n_step, n_hidden * num_directions(=2)], F matrix
    def attention_net(self, lstm_output, final_state):
        hidden = final_state.view(-1, n_hidden * 2, 1)   # hidden : [batch_size, n_hidden * num_directions(=2), 1(=n_layer)]
        attn_weights = torch.bmm(lstm_output, hidden).squeeze(2) # attn_weights : [batch_size, n_step]
        soft_attn_weights = F.softmax(attn_weights, 1)
&