Yukta/rl module finish#127

Open

yuktakul04 wants to merge 37 commits into

google:copybara_pushfrom

yuktakul04:yukta/rl-module-finish

Commits on Oct 17, 2025

feat: reinforcement learning PR#2; several additions/improvements to rl pipeline

gabriel-trigo
authored and
yuktakul04
committed
Update pyproject.toml

s2t2
authored and
yuktakul04
committed
fix: fix linting errors of previous commit

gabriel-trigo
authored and
yuktakul04
committed
Update PR Template

s2t2
authored and
yuktakul04
committed
Update PR Template

s2t2
authored and
yuktakul04
committed
Restore original formatting

s2t2
authored and
yuktakul04
committed
Restore original formatting

s2t2
authored and
yuktakul04
committed
Clean top of files

s2t2
authored and
yuktakul04
committed
Refactor filepaths

s2t2
authored and
yuktakul04
committed
Refactor filepaths

s2t2
authored and
yuktakul04
committed
Refactor and test temp conversion functions; closes #25

s2t2
authored and
yuktakul04
committed
Refactor temp conversion tests

s2t2
authored and
yuktakul04
committed
Review eval script

s2t2
authored and
yuktakul04
committed
Remove redundant variable setting

s2t2
authored and
yuktakul04
committed
Fix failing test

s2t2
authored and
yuktakul04
committed
Repro generate configs script; use absl flags because argparse not working

s2t2
authored and
yuktakul04
committed
Update gitignore

s2t2
authored and
yuktakul04
committed
Test config file generation

s2t2
authored and
yuktakul04
committed
Test read config file

s2t2
authored and
yuktakul04
committed
Fix file names - remove quote

s2t2
authored and
yuktakul04
committed
Describe the config generation script

s2t2
authored and
yuktakul04
committed
Flags WIP

s2t2
authored and
yuktakul04
committed
Attempt to reproduce starter buffer script; fix #115

s2t2
authored and
yuktakul04
committed
Test starter buffer population

s2t2
authored and
yuktakul04
committed
Refactor test: use setup, teardown, and temp dir

s2t2
authored and
yuktakul04
committed
WIP - reproduce train script, run into known issue

s2t2
authored and
yuktakul04
committed
Hotfix known issue

s2t2
authored and
yuktakul04
committed
Generate example starter buffers for training and testing

s2t2
authored and
yuktakul04
committed
WIP - refactor and test RL agent trainer

s2t2
authored and
yuktakul04
committed
Regenerate starter buffer for testing

s2t2
authored and
yuktakul04
committed
Decrease number of training steps when testing

s2t2
authored and
yuktakul04
committed
WIP - reproducing eval script - encounter env config errors

s2t2
authored and
yuktakul04
committed
Reproduce eval script

s2t2
authored and
yuktakul04
committed
WIP - refactor eval script; need to save schedule policy results charts as well

s2t2
authored and
yuktakul04
committed
feat(rl): fix replay buffer integration, add seeding & tests, and harden paths
yuktakul04
committed
fix(replay): make dm-reverb optional; fallback to TFUniform on macOS/py3.11
yuktakul04
committed
fix(replay): TFUniform fallback with batched observer; dm-reverb optional
yuktakul04
committed