Fix : add gradient w.r.t. exponent in matrix_power backward

[Emma5938]

Context

This PR fixes issue #25 regarding the backward behavior of matrix_power, where the gradient with respect to the exponent parameter was previously not implemented.

Changes

This PR updates the backward function of matrix_power to include a gradient computation w.r.t. the exponent parameter.

Tests

Running pytest locally results in 8 failing tests.
These failures are also present on the current main branch, indicating they are unrelated to this PR.

Authors

• @Emma5938
• @Marco-Congedo
• @isacostamaia
• SAID Salem

Affiliation

Gipsa-lab, University of Grenoble Alpes, CNRS, Grenoble INP

[bruAristimunha]

Hey @Emma5938!
Many thanks for the contribution ;)

Looks good for me, I could not see the tests that are failing in the CI. I was wondering, could you please list for us?

I will fix the documentation issue soon, I think it is some small transiting download problem.

[Emma5938]

Hey! Thanks for the review.

The failing tests when running pytest locally are:

• FAILED tests/test_integration.py::test_module_dtype[WaveletConv-float32-cuda] - RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same or input should be a MKLDNN tensor and weight is a dense tensor

• FAILED tests/test_integration.py::test_module_dtype[WaveletConv-float64-cuda] - RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument weight in method wrapper_CUDA___slow_conv2d_forward)

• FAILED tests/test_integration.py::test_module_dtype[WaveletConv-complex64-cuda] - RuntimeError: Input type (CPUComplexFloatType) and weight type (CUDAComplexFloatType) should be the same

• FAILED tests/test_integration.py::test_module_dtype[WaveletConv-complex128-cuda] - RuntimeError: Input type (CPUComplexDoubleType) and weight type (CUDAComplexDoubleType) should be the same

• FAILED tests/test_modules.py::test_bimap_instantiation_and_shape[Kaiming (Parametrized)] - AssertionError: Failed device check for Kaiming (Parametrized)

• FAILED tests/test_modules.py::test_bimap_instantiation_and_shape[Orthogonal (Parametrized)] - AssertionError: Failed device check for Orthogonal (Parametrized)

• FAILED tests/test_modules.py::test_bimap_instantiation_and_shape[Stiefel (Non-Param)] - AssertionError: Failed device check for Stiefel (Non-Param)

• FAILED tests/test_modules.py::test_bimap_instantiation_and_shape[Dim Increase Kaiming (Param)] - AssertionError: Failed device check for Dim Increase Kaiming (Param)

Most of them seem related to CUDA/device mismatch issues.

[bruAristimunha]

Many thanks for the contribution @Emma5938, @Marco-Congedo and @isacostamaia.