Fix gmres errors in CI triggered by new Scipy version #81
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
Scipy GMRES now modifies the preconditioned residual in place. This is perfectly reasonable, but we were passing in jax numpy arrays, which do not allow this. This was unsafe, but we got away with it. This commit makes deep copies of jax arrays to plain numpy arrays in the operators and initial data that the scipy gmres sees.
The relative tolerance for convergence was being set to a fraction of the rhs norm, but the `tol` option is already a relative tolerance. This commit also fies a depracation warning, since the `tol` parameter is deprecated in facor of `rtol`.
tupek2
approved these changes
Feb 27, 2024
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Outdated suggestions cannot be applied.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.
We were passing JAX arrays to Scipy GMRES and duck typing was making it work. However, Scipy's GMRES has been refactored, and now it tries to modify some arrays in place. This is perfectly reasonable, but now the immutability of JAX arrays is causing a problem. I fixed this by making deep copies of the data to plain numpy arrays (including output of the matvec and preconditioner operators).
I also did some small maintenance in the nearby code. Please take a look at the change in the relative tolerance specification. I think it fixes a conceptual error, but it's going to cause a change in behavior. This modification also has the benefit of eliminating a deprecated parameter (Scipy is phasing out
tolin favor of explicitrtolin the iterative solvers.)Fixes #78