70e4d7fe60a9658bb27b9f5fb67592a1222b2ec3,spotlight/sequence/representations.py,PoolNet,user_representation,#PoolNet#Any#,23

Before Change

        // Average representations, ignoring padding.

        sequence_embedding_sum = (sequence_embeddings
                                  .sum(1)
                                  .view(item_sequences.size()[0], -1))
        non_padding_entries = ((item_sequences != PADDING_IDX)
                               .float()
                               .sum(1)
                               .expand_as(sequence_embedding_sum))

        user_representations = (
            sequence_embedding_sum / (non_padding_entries + 1)
        )

        return user_representations

    def forward(self, user_representations, targets):

        target_embedding = self.item_embeddings(targets)

After Change

    def user_representation(self, item_sequences):

        // Make the embedding dimension the channel dimension
        sequence_embeddings = (self.item_embeddings(item_sequences)
                               .permute(0, 2, 1))

        // Add a trailing dimension of 1
        sequence_embeddings = (sequence_embeddings
                               .unsqueeze(3))

        // Pad it with zeros from left
        sequence_embeddings = F.pad(sequence_embeddings,
                                    (0, 0, 1, 0))

        // Average representations, ignoring padding.
        sequence_embedding_sum = torch.cumsum(sequence_embeddings, 2)
        non_padding_entries = (
            torch.cumsum((sequence_embeddings != 0.0).float(), 2)
            .expand_as(sequence_embedding_sum)
        )

        user_representations = (
            sequence_embedding_sum / (non_padding_entries + 1)
        ).squeeze(3)

        return user_representations[:, :, :-1], user_representations[:, :, -1]

    def forward(self, user_representations, targets):

Instances