[lenianiva]

This fixes a bug where during backpropagation the size of an array e.g. (1, 1, 0) does not match the size of the gradient e.g. (4, 1, 0). The error only happens for zero-sized arrays.

See an example here: JuliaGraphs/GraphNeuralNetworks.jl#638

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.63%. Comparing base (0ae32d8) to head (e2b5670).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #702   +/-   ##
=======================================
  Coverage   93.63%   93.63%           
=======================================
  Files          15       15           
  Lines        1022     1022           
=======================================
  Hits          957      957           
  Misses         65       65           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[CarloLucibello]

merge?

[devmotion]

It would be interesting to know first why the gradient has an incorrect shape. And ideally then we could fix the underlying issue instead of relaxing the projection.

[lenianiva]

It would be interesting to know first why the gradient has an incorrect shape. And ideally then we could fix the underlying issue instead of relaxing the projection.

After poking around for months I still cannot find the underlying issue that caused this.

[devmotion]

Were you able to figure out the underlying issue/path that caused this gradient shape?

[lenianiva]

Were you able to figure out the underlying issue/path that caused this gradient shape?

The bug seems to have been fixed on main.