Feature/low rank approx #365
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…r low rank approximation solver
mdbenito
added
the
new-method
…to remove dependency of conftest on torch
…eration number due to randomly observed situtation, where LISSA does not converge do desired error tolerance
…environment, in order to not invoke tests from other directories like /notebooks
…notebooks environemnt dependendencies
… docs build" This reverts commit ea8774e.
mdbenito
reviewed
Jun 28, 2023
* remove parameter x0 * simplify if/else condition
Feature/344 functorch
Avoid confusion with TensorType defined in influence.frameworks.twice_differentiable, which serves a different purpose
* harmonize docstring format * add unit test for align_structure * move import to toplevel * simplify path existence check
This was
linked to
issues
Jul 26, 2023
…d torch environment, in order to not invoke tests from other directories like /notebooks" This reverts commit 5a28716.
AnesBenmerzoug
approved these changes
Jul 28, 2023
|
@schroedk There is a bug in Github with using [skip ci] as the last commit before a merge. It prevents CI from running and because CI needs to be successful in order to merge the PR that means we're stuck in a deadlock. We can either push a dummy commit (not really recommended) or just override the merge requirements (if you can't do it then just tell and I can do it for you). |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
This PR adds a new solver to compute influence functions. It is based on a low rank approximation, via approximate eigen decompositions. It calls ARPack routines via scipy, which seems to be good thing to do, in comparison to implement the Arnoldi/Lanczos in (cpu) torch.
Considerations to still mark this as draft:
I followed the existing abstraction (TorchTwiceDifferentiable), but would prefer to get rid of it, as it seems quite inflexible to me, also there is some broken signature, as the type hint suggest to follow the base class TwiceDifferentiable, while it actually needs TorchTwiceDifferentiable. I will create an issue for this. Should this be in this PR? I prefer to do it an a further PRseparate PROnce we streamlined the interfaces, I we could add support for JAX as well. This will be a large PR, I guessseparate PRThis PR closes #282
Changes
Checklist
"nbsphinx":"hidden"