7570ee1d758c93dd4054009fc12be6cc31b54031,gpytorch/lazy/interpolated_lazy_tensor.py,InterpolatedLazyTensor,_get_indices,#InterpolatedLazyTensor#Any#Any#,261

Before Change

        interp_values = torch.matmul(left_interp_values, right_interp_values)

        // Batch compute values that will be non-zero for row k
        if left_interp_indices.ndimension() == 3:
            left_interp_indices = left_interp_indices.unsqueeze(-1).expand(n_data, n_interp, n_interp).contiguous()
            right_interp_indices = right_interp_indices.unsqueeze(-2).expand(n_data, n_interp, n_interp).contiguous()
        else:
            left_interp_indices = left_interp_indices.unsqueeze(-1).expand(n_data, n_interp, n_interp).contiguous()
            right_interp_indices = right_interp_indices.unsqueeze(-2).expand(n_data, n_interp, n_interp).contiguous()
        base_var_vals = self.base_lazy_tensor._get_indices(left_interp_indices.view(-1), right_interp_indices.view(-1))
        base_var_vals = base_var_vals.view(left_interp_indices.size())
        res = (interp_values * base_var_vals).sum(-1).sum(-1)
        return res

After Change

        left_interp_indices = self.left_interp_indices.__getitem__((*batch_indices, left_indices))
        left_interp_values = self.left_interp_values.__getitem__((*batch_indices, left_indices))
        right_interp_indices = self.right_interp_indices.__getitem__((*batch_indices, right_indices))
        right_interp_values = self.right_interp_values.__getitem__((*batch_indices, right_indices))

        n_data, n_interp = left_interp_indices.size()

Instances