[tune] fix non-deterministic category sampling by switching back to np.random.choice
#13710
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Outdated suggestions cannot be applied.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.
Why are these changes needed?
Sampling categories in Ray Tune was non-deterministic even when setting a seed. It is unclear where this comes from exactly, but it might be some external library using python's
random
module. Switching backtune.choice
sampling tonp.random.choice
fixes this issue.Note that we introduced the switch to
random.choice
in #9451 because of typing incompatibilities. However, transforming the sampled choices to a list should fix these issues. An alternative would be to sample indices and return a list of these indices, but we'll try this regular route as it should be more efficient.Related issue number
Closes #13295
Original issue for switching from
np.random.choice
torandom.choice
: #8976Related PR: #9451
Checks
scripts/format.sh
to lint the changes in this PR.