bbfbfce5c8c18b39d541d912ffb11182f7d06dfd,allennlp/modules/similarity_functions/linear.py,LinearSimilarity,forward,#LinearSimilarity#Any#Any#,58

Before Change

        else:
            view_args = [-1] + list(combined_tensors.size()[-2:])
            reshaped_tensor = combined_tensors.view(*view_args)
            unsqueezed_weight = self._weight_vector.unsqueeze(1).unsqueeze(0)
            reshaped_weight = unsqueezed_weight.expand(reshaped_tensor.size()[0],
                                                       self._weight_vector.size()[0],
                                                       1)
            reshaped_dot_product = reshaped_tensor.bmm(reshaped_weight)
            view_args = combined_tensors.size()[:-1]
            dot_product = reshaped_dot_product.view(*view_args)
        return self._activation(dot_product + self._bias.expand_as(dot_product)).squeeze(dim=-1)

After Change

    @overrides
    def forward(self, tensor_1: torch.Tensor, tensor_2: torch.Tensor) -> torch.Tensor:
        combined_tensors = self._combine_tensors(tensor_1, tensor_2)
        dot_product = torch.matmul(combined_tensors, self._weight_vector)
        return self._activation(dot_product + self._bias)

    def _combine_tensors(self, tensor_1: torch.Tensor, tensor_2: torch.Tensor) -> torch.Tensor:
        combined_tensor = self._get_combination(self._combinations[0], tensor_1, tensor_2)

Instances