Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[ADD] Change KFAC and KFAC inverse to purely PyTorch #149

Draft
wants to merge 36 commits into
base: main
Choose a base branch
from
Draft

[ADD] Change KFAC and KFAC inverse to purely PyTorch #149

wants to merge 36 commits into from

Conversation

f-dangel
Copy link
Owner

@f-dangel f-dangel commented Oct 27, 2024 •
edited
Loading

Review after #142, #145, #146, #147, #148, #149.

@f-dangel
[ADD] Minimal linear operator interface for PyTorch
06ebbf2
@f-dangel
[FIX] Linters
f652a02
@f-dangel
[ADD] Replicate current base class but inherit from `PyTorchLinearOpe…
9fa1b6a 
…rator`
@f-dangel
[DEL] Unused import
62dc997
@f-dangel
[ADD] Implement Hessian as CurvatureLinearOperator
59c6369
@f-dangel
[REF] Combine full and block-diagonal matrix multiply tests
07144a3
@f-dangel
[ADD] Automatically implement adjoint for self-adjoint ops
9c538ea
@f-dangel
[ADD] Implement GGN as CurvatureLinearOperator
3a8ac95
@f-dangel
[FIX] flake8
73cdb5a
@f-dangel
[FIX] Increase tolerance to make test pass
e8f77e8
@f-dangel
[DEL] Unused import
8b6fed7
@f-dangel
[ADD] Implement FisherMCLinearOperator as CurvatureLinearOperator
0ed579f
@f-dangel
[ADD] Implement Jacobians as CurvatureLinearOperator
d72094b
@f-dangel
[REF] Extract error test with MutableMapping and ``batch_size_fn=…
85e6c6b 
…None``
@f-dangel
[ADD] Class attribute for checking deterministic batches
c819189
@f-dangel
[ADD] Extract getters for _in_shape and _out_shape
d38bb4f
@f-dangel
[ADD] Test deterministic batches
4f546fc
@f-dangel
[DOC] Add documentation and type annotations
87c6eba
@f-dangel
[ADD] Implement empirical Fisher as CurvatureLinearOperator
c6fe684
@f-dangel
Merge branch 'main' into fisher-linop
eba5976
@f-dangel
[REF] Extract testing matrix multiplies in expectation
e678ea2
@f-dangel
[FIX] Examples
32523ad
@f-dangel
[ADD] Check deterministic batches in MC-Fisher (fixes #75)
9ff958b
@f-dangel
Merge branch 'jacobian-linop' into fisher-linop
76a5190
@f-dangel
Merge branch 'ef-linop' into act-hessian-linop
f67ca0f
@f-dangel
[ADD] Implement activation Hessian as CurvatureLinearOperator
9be72fc
@f-dangel
[ADD] Implement KFAC as CurvatureLinearOperator
a9ea3fc
@f-dangel
[DEL] Remove original base class
012cc21
@f-dangel
[FIX] Darglint
4d0020c
@f-dangel
[DEL] Remove unused imports
efaf74a
@f-dangel f-dangel marked this pull request as draft October 27, 2024 16:26
@f-dangel f-dangel changed the title [ADD] Change empirical Fisher and activation Hessian to purely PyTorch [ADD] Change KFAC and KFAC inverse to purely PyTorch Oct 27, 2024
@coveralls
Copy link

Pull Request Test Coverage Report for Build 11542141176

Details

  • 363 of 390 (93.08%) changed or added relevant lines in 10 files are covered.
  • 51 unchanged lines in 3 files lost coverage.
  • Overall coverage decreased (-2.0%) to 89.621%
Changes Missing Coverage Covered Lines Changed/Added Lines %
curvlinops/fisher.py 5 10 50.0%
curvlinops/_torch_base.py 198 220 90.0%
Files with Coverage Reduction New Missed Lines %
curvlinops/inverse.py 1 90.36%
curvlinops/examples/utils.py 10 31.25%
curvlinops/fisher.py 40 25.76%
Totals Coverage Status
Change from base Build 11523760748: -2.0%
Covered Lines: 1347
Relevant Lines: 1503
💛 - Coveralls

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
Linear operators for PyTorch and SciPy
Development

Successfully merging this pull request may close these issues.
2 participants
@f-dangel @coveralls