patrick-kidger / patrick-kidger/diffrax

Have `ImplicitAdjoint` use `lstsq` during jvp

Open
#137 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

bug next
Dominant language
Python
Stars
2.1k
Forks
189
Avg merge
3d 18h
Merged PRs (30d)
1

Description

At the moment it uses jnp.linalg.solve, which assumes that the solve operation is well-posed. This needn't actually be the case, though, and at the moment the gradient we obtain could be offset by anything in the kernel of the Jacobian.

(This has been fine for our other main usage of implicit differentiation -- implicit diffeq solvers -- since these are well-posed by construction.)

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Locate the ImplicitAdjoint jvp implementation and the current jnp.linalg.solve call. Read how the Jacobian's kernel can affect the gradient, then verify the change with a rank-deficient Jacobian case and confirm that the resulting gradient uses the least-squares solve.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
backend
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.