pymc-devs / pymc-devs/pytensor

Implement ScalarLoop grad

Open
#300 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

gradients
Dominant language
Python
Stars
644
Forks
208
Avg merge
2d 14h
Merged PRs (30d)
16

Description

@aseyboldt mentioned we should be able to implement the gradient as another ScalarLoop, via forward auto-diff, which in this case should be equivalent to backward-autodiff.

It should also be possible to fuse the loops, but hopefully that's one thing the compilers can easily figure out, once they get merged inside the same Composite?

Originally posted by @ricardoV94 in https://github.com/pymc-devs/pytensor/issues/283#issuecomment-1539706699

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

Start by locating the ScalarLoop implementation and its forward- and backward-autodiff paths, then inspect how loops are combined inside Composite. The work is complete when the gradient can be represented as another ScalarLoop and the existing autodiff behavior remains equivalent.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
compilers
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.