Add Ops for LU Factorization #1218
Draft
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
ricardoV94
force-pushed
the
LU-factorization
branch
from
3ceece7 to
e57bf39
Compare
jessegrabowski
force-pushed
the
LU-factorization
branch
3 times, most recently
from
b1f8c9d to
2a5b361
Compare
…ontinuous output types
jessegrabowski
force-pushed
the
LU-factorization
branch
from
2a5b361 to
93314a5
Compare
ricardoV94
reviewed
Feb 28, 2025
|
|
||
|
|
||
| def lu( | ||
| a: TensorLike, permute_l=False, check_finite=True, p_indices=False |
There was a problem hiding this comment.
you can allow overwrite_a in the signature but just ignore it like JAX does
| compare_numba_and_py([A], out, test_inputs=[A_val], inplace=True) | ||
|
|
||
| else: | ||
| # compare_numba_and_py fails: NotImplementedError: Non-jitted BlockwiseWithCoreShape not implemented |
There was a problem hiding this comment.
just pass eval_obj_mode=False
| def __init__(self, *, overwrite_a=False, check_finite=True): | ||
| self.overwrite_a = overwrite_a | ||
| self.check_finite = check_finite | ||
| self.gufunc_signature = "(m,m)->(m,m),(m)" |
There was a problem hiding this comment.
Put signature out of init since it's constant
| f = pytensor.function([A], out, mode="NUMBA") | ||
|
|
||
| if len(shape) == 2: | ||
| compare_numba_and_py([A], out, [A_val], inplace=True) |
There was a problem hiding this comment.
Same you can use compare_numba_and_py
|
Any benchmarks on solve written with these Ops? |
|
Working on that next, just ironing out some bugs in the |
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Outdated suggestions cannot be applied.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 will add the following Ops:
As well as dispatches for numba/jax (and maybe torch, though help is welcome there).
The reason for wanting these is that it will make the gradients of solve faster. I think this is a major reason why jax has faster gradients than us (at least when solve is implicated). They route everything to
lu_solve(lu_factor(A), b), and reuselu_factor(A)in the backward pass.Related Issue
Checklist
Type of change
📚 Documentation preview 📚: https://pytensor--1218.org.readthedocs.build/en/1218/