Backpropagation for sparse matrix indexing is problematic (colab provided) #45996
Labels
module: sparse
Related to torch.sparse
triaged
This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Projects
馃悰 Bug
When indexing the sparse matrix, it performs well in the forward, but in backpropagation, the gradients become all zeros (i.e. empty sparse matrix).
To Reproduce
I put up a toy case in colab to reproduce.
The strange behavior is the gradients (in sparse) is empty.
Expected behavior
non-zero gradients appear at non-zero entry location.
cc @vincentqb @aocsa @nikitaved @pearu @mruberry
The text was updated successfully, but these errors were encountered: