MCPcopy Create free account
hub / github.com/Tencent/NeuralNLP-NeuralClassifier / forward

Method forward

model/classification/dpcnn.py:61–79  ·  view source on GitHub ↗
(self, batch)

Source from the content-addressed store, hash-verified

59 return params
60
61 def forward(self, batch):
62 if self.config.feature.feature_names[0] == "token":
63 embedding = self.token_embedding(
64 batch[cDataset.DOC_TOKEN].to(self.config.device))
65 else:
66 embedding = self.char_embedding(
67 batch[cDataset.DOC_CHAR]).to(self.config.device)
68 embedding = embedding.permute(0, 2, 1)
69 conv_embedding = self.convert_conv(embedding)
70 conv_features = self.convs[0](conv_embedding)
71 conv_features = conv_embedding + conv_features
72 for i in range(1, len(self.convs)):
73 block_features = F.max_pool1d(
74 conv_features, self.kernel_size, self.pooling_stride)
75 conv_features = self.convs[i](block_features)
76 conv_features = conv_features + block_features
77 doc_embedding = F.max_pool1d(
78 conv_features, conv_features.size(2)).squeeze()
79 return self.dropout(self.linear(doc_embedding))

Callers

nothing calls this directly

Calls

no outgoing calls

Tested by

no test coverage detected