MCPcopy Create free account
hub / github.com/NJUNLP/GTS / forward

Method forward

code/NNModel/model.py:162–183  ·  view source on GitHub ↗
(self, x, x_len, x_mask)

Source from the content-addressed store, hash-verified

160 return logits_list
161
162 def forward(self, x, x_len, x_mask):
163 x_emb = torch.cat((self.gen_embedding(x), self.domain_embedding(x)), dim=2)
164 x_emb = self.dropout(x_emb).transpose(1, 2)
165 x_conv = torch.nn.functional.relu(torch.cat((self.conv1(x_emb), self.conv2(x_emb)), dim=1))
166 x_conv = self.dropout(x_conv)
167 x_conv = torch.nn.functional.relu(self.conv3(x_conv))
168 x_conv = self.dropout(x_conv)
169 x_conv = torch.nn.functional.relu(self.conv4(x_conv))
170 x_conv = self.dropout(x_conv)
171 x_conv = torch.nn.functional.relu(self.conv5(x_conv))
172 x_conv = x_conv.transpose(1, 2)
173 x_conv = x_conv[:, :x_len[0], :]
174
175 feature_attention = self.attention_layer.forward_perceptron(x_conv, x_conv, x_mask[:, :x_len[0]])
176 x_conv = x_conv + feature_attention
177
178 x_conv = x_conv.unsqueeze(2).expand([-1, -1, x_len[0], -1])
179 x_conv_T = x_conv.transpose(1, 2)
180 features = torch.cat([x_conv, x_conv_T], dim=3)
181
182 logits = self.multi_hops(features, x_len, x_mask, self.args.nhops)
183 return [logits[-1]]
184

Callers

nothing calls this directly

Calls 2

multi_hopsMethod · 0.95
forward_perceptronMethod · 0.80

Tested by

no test coverage detected