model.py 文件源码

def forward(self, x):
        n_idx = 0
        c_idx = 1
        h_idx = 2
        w_idx = 3

        x = self.lookup_table(x)
        x = x.unsqueeze(c_idx)

        enc_outs = []
        for encoder in self.encoders:
            enc_ = F.relu(encoder(x))
            k_h = enc_.size()[h_idx]
            enc_ = F.max_pool2d(enc_, kernel_size=(k_h, 1))
            enc_ = enc_.squeeze(w_idx)
            enc_ = enc_.squeeze(h_idx)
            enc_outs.append(enc_)

        encoding = self.dropout(torch.cat(enc_outs, 1))
        return F.log_softmax(self.logistic(encoding))