Conversation
- Add ard=False to bayes_search() in all six policy classes (discrete, range, discrete_multi, range_multi, discrete_unified, range_unified) - Introduce _make_gp_predictor(num_dim=None) in discrete and range policies to create GP predictor with optional ARD (Automatic Relevance Determination) - When ard=True, use Gauss kernel with per-dimension width; require training data (e.g. from random_search) before bayes_search for input dimension - Add test_bayes_search_ard to verify ard=True sets predictor with ARD kernel Made-with: Cursor
- Add ard=False parameter to bayes_search in discrete, range, discrete_multi, range_multi policies - Add get_kernel_length_scale() to policy for inspecting learned kernel width(s) - Add ard_example.py: compare ard=True vs ard=False (best value, length scale, permutation importance) - Add unit test for get_kernel_length_scale Made-with: Cursor
…model - Add _make_gp_model() to discrete and range policies; use it for both GP and BLM. - When num_rand_basis > 0 and ard=True, pass ARD GP model to blm_predictor so random feature expansion uses per-dimension length scales. - _make_gp_model(ard=False) now returns the default GP model instead of None. - _make_gp_predictor always uses the model from _make_gp_model. - Apply BLM+ARD init in discrete_multi, range_multi, discrete_unified, range_unified. - Simplify ard and get_kernel_length_scale docstrings. Made-with: Cursor
Contributor
There was a problem hiding this comment.
Pull request overview
This PR adds Automatic Relevance Determination (ARD) support to PHYSBO's Bayesian optimization framework. ARD enables per-dimension length scales in the Gaussian kernel, allowing the GP surrogate to automatically identify which input dimensions are relevant to the objective function.
Changes:
- Added
ard=Trueparameter tobayes_searchmethod across all 6 policy classes (discrete, discrete_multi, discrete_unified, range, range_multi, range_unified) - Added
get_kernel_length_scale()method to retrieve learned length scales from the GP model in all policy variants - Added
_make_gp_modeland_make_gp_predictorhelper methods to create GP models/predictors with optional ARD kernel - Added documentation (English/Japanese tutorials), example script, and unit tests for the new functionality
Reviewed changes
Copilot reviewed 12 out of 12 changed files in this pull request and generated 5 comments.
Show a summary per file
| File | Description |
|---|---|
src/physbo/search/discrete/_policy.py |
Core ARD support: ard param in bayes_search, get_kernel_length_scale, _make_gp_model, _make_gp_predictor |
src/physbo/search/range/_policy.py |
Mirror of discrete's ARD support for continuous search space policies |
src/physbo/search/discrete_multi/_policy.py |
ARD in multi-objective discrete search; overrides get_kernel_length_scale for list return |
src/physbo/search/range_multi/_policy.py |
ARD in multi-objective range search; overrides get_kernel_length_scale for list return |
src/physbo/search/discrete_unified/_policy.py |
ARD in unified-objective discrete search (inherits get_kernel_length_scale) |
src/physbo/search/range_unified/_policy.py |
ARD in unified-objective range search (inherits get_kernel_length_scale) |
tests/unit/test_policy.py |
Unit tests for ARD kernel creation and get_kernel_length_scale (discrete only) |
examples/single_objective/ard_example.py |
Example script demonstrating ARD vs isotropic kernel |
docs/sphinx/manual/en/source/notebook/tutorial_ard.ipynb |
English ARD tutorial notebook |
docs/sphinx/manual/ja/source/notebook/tutorial_ard.ipynb |
Japanese ARD tutorial notebook |
docs/sphinx/manual/en/source/notebook/index.rst |
Add ARD tutorial to English docs index |
docs/sphinx/manual/ja/source/notebook/index.rst |
Add ARD tutorial to Japanese docs index |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
- Add Model.create_default(ard=False, num_dim=None) in gp.core._model to build default Gauss kernel + Const mean + Gauss likelihood. - Remove _make_gp_model from discrete and range policies; use gp.core.Model.create_default(ard=..., num_dim=...) instead. - Update discrete_multi, range_multi, discrete_unified, range_unified to call Model.create_default for BLM+ARD predictor init. - Add gp import where needed in multi/unified policy modules. Made-with: Cursor
Contributor
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 13 out of 13 changed files in this pull request and generated 9 comments.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
- range: remove num_dim from _init_predictor and _make_gp_predictor; add _make_blm_predictor(); both use self.ard and self.get_num_dim(). - range_multi, range_unified: create BLM predictor via _make_blm_predictor() instead of inline create_default; drop unused gp import. Made-with: Cursor
Contributor
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 13 out of 13 changed files in this pull request and generated 7 comments.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
This file contains hidden or 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
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.
Instead of the Gaussian kernel
the ARD kernel
is now available by passing
ard=Truetobayes_searchmethod of eachPolicy.A new method$\vec{\eta}$ , which reflects sensiblity of each dimension of input to the objective function.
policy.get_kernel_length_scalereturns