Fix bug in optimize_objective with fixed features
#2691
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D68464825 |
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #2691 +/- ##
=======================================
Coverage 99.98% 99.98%
=======================================
Files 200 200
Lines 18232 18238 +6
=======================================
+ Hits 18230 18236 +6
Misses 2 2 ☔ View full report in Codecov by Sentry. |
Summary: **Context:** As per pytorch#2686, bounds for `optimize_acqf` are not constructed correctly in `optimize_objective`, which is used in input constructors for qKG-type acquisition functions. This issue wasn't surfaced by unit tests because `optimize_acqf` was mocked out. In the process of shoring up the test, I discovered a second bug: This `optimize_objective` doesn't work with constraints, because the optimizer is set to be L-BFGS-B when it isn't otherwise specified, and L-BFGS-B doesn't work with BoTorch-style constraints (only simple box constraints, aka BoTorch bounds). So I guess the input constructors for qKG-style acquisition functions haven't been working with fixed features or with constraints for a long time -- both usages would just error. The existing unit test should have caught this but didn't due to use of mocks, so I removed the mocking. **Changes:** In `optimize_objective`: * Use `bounds.shape` instead of `len(bounds)` when constructing a list of features for `fixed_features_list` * Don't specify 'method' if the user doesn't pass it, so it can be automatically chosen based on the presence of constraints. Other: * In `optimize_acqf`, cleaned up some logic. This doesn't have any effect on behavior. * Added a type annotation Differential Revision: D68464825
esantorella
force-pushed
the
export-D68464825
branch
from
e82b5d8 to
7b8d92c
Compare
|
This pull request was exported from Phabricator. Differential Revision: D68464825 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D68464825 |
|
This pull request has been merged in 6b75672. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Summary:
Context: As per #2686, bounds for
optimize_acqfare not constructed correctly inoptimize_objective, which is used in input constructors for qKG-type acquisition functions. This issue wasn't surfaced by unit tests becauseoptimize_acqfwas mocked out. In the process of shoring up the test, I discovered a second bug: Thisoptimize_objectivedoesn't work with constraints, because the optimizer is set to be L-BFGS-B when it isn't otherwise specified, and L-BFGS-B doesn't work with BoTorch-style constraints (only simple box constraints, aka BoTorch bounds).So I guess the input constructors for qKG-style acquisition functions haven't been working with fixed features or with constraints for a long time -- both usages would just error.
The existing unit test should have caught this but didn't due to use of mocks, so I removed the mocking.
Changes:
In
optimize_objective:bounds.shapeinstead oflen(bounds)when constructing a list of features forfixed_features_listOther:
optimize_acqf, cleaned up some logic. This doesn't have any effect on behavior.Differential Revision: D68464825