EPIG implementation - Githubissues

baal-org / baal

Bayesian active learning library for research and industrial usecases.

https://baal.readthedocs.io

Apache License 2.0

862 stars 86 forks source link

EPIG implementation #270

Closed reeshipaul closed 4 months ago

reeshipaul commented 1 year ago

Summary:

Issue #259

Features:

Checklist:

[ ] Your code is documented (To validate this, add your module to tests/documentation_test.py).
[ ] Your code is tested with unit tests.
[x] You moved your Issue to the PR state.

Dref360 commented 5 months ago

With @fbickfordsmith latest AISTAT paper link I think we should really more this PR forward. I refactored a bit and assuming the equations in EPIG are correct I think we can merge that pretty soon.

Freddie, if your energy takes you there, could you review the code of EPIG? In particular the compute_score function which receives both unlabelled probabilities and training probabilities. It returns the EPIG score for all unlabelled items.

fbickfordsmith commented 5 months ago

Great that you're moving this forward! Happy to support. I'm at ICLR this week so can't look at the code for a while, but I've added this to my to-do list.

fbickfordsmith commented 4 months ago

Nice work @Dref360 @reeshipaul! Looks good overall.

I have some minor suggestions. Probably easiest if I push some changes for you to look at.

What's the best way to do that? If @reeshipaul gives me push access to https://github.com/reeshipaul/baal then I could create a branch there. Open to alternatives too.

Dref360 commented 4 months ago

In case @reeshipaul doesn't answer back, feel free to submit comments with suggestion. If by the end of the week we don't hear back, I'll merge this branch in a feature branch and you can fork baal-org/baal directly.

Dref360 commented 4 months ago

@fbickfordsmith this has been merged into baal-org:features/epig_release, please submit a PR against it with your changes.