def forward()

in src/model.py [0:0]

• 23 lines of code
• 3 McCabe index (conditional complexity)

    def forward(self, input):
        """ Appends meta data embeddings to input and passes through LSTM."""
        input_ids = input["input"].long()
        input_lens = input["text_len"]
        input_embs = self.embeddings(input_ids)

        if self.use_md:
            md_input = self.metadata_constructor.preprocess_md(input["md"], self.embeddings)
            md_emb = self.metadata_constructor(md_input)

            # Prepending meta data information to text
            sos_embs = input_embs[:, :1, :]
            text_embs = input_embs[:, 1:, :]

            joined_embs = torch.cat((sos_embs, md_emb, text_embs), dim=1)
            joined_lens = 1 + input_lens # fixed metadata embedding

            packed_input = pack_padded_sequence(joined_embs, joined_lens,
                                                batch_first=True, enforce_sorted=False)
        else:
            packed_input = pack_padded_sequence(input_embs, input_lens,
                                                batch_first=True, enforce_sorted=False)

        # output: batch, seq_len, hidden_size
        lstm_out, _ = self.lstm(packed_input)
        lstm_out, _ = pad_packed_sequence(lstm_out, batch_first=True)

        if self.use_weight_tying:
            vocab_predictions = self.vocab_projection(self.embedding_projection(lstm_out))
        else:
            vocab_predictions = self.vocab_projection(lstm_out)

        return vocab_predictions