Python and C++ code refactoring

[carloslihu]

• Python code is now formatted with black and isort, and has been refactored according to PEP 8 style guides.
• Python code partially commented with google docstring format.
• C++ code partially commented with doxygen docstring format.
• Scott's and Normal Reference Rule's bandwidth calculation have been reordered and commented.
• ArcOperatorSet::update_incoming_arcs_scores formulas have been reordered and commented.

[carloslihu]

Hello David, thank you for reviewing my changes, they are a lot:

• The bug I'm talking about is about performing HC with cross-validation when you have a variable with low variability (but not 0 variance). For instance, a variable A with training values [0, 0, ..., 0, 1].
  When performing the validated k-fold cross-validation, it finds problems fitting the distribution when the "1" value is not found in the training fold.
  It is not a bug, more like a problem I found with some datasets. I do not have a clear solution for the algorithm. You may ignore the comment; I will remove it.

• I still have pending to review that the python tests are still working and cleaning comments.
  It's a minor change, but I have planned to finish this month.
  Please wait for them, thanks

[carloslihu]

Another issue is that I have retrocompatibility with the tests until the tag v0.5.2
However, afterwards I started using for KDE by default the diagonal covariance as the bandwidth matrix, instead of the full covariance (as suggested here)

This KDE learning faster and removes the Singular Covariance Matrix error.
Although it may return less accurate Bandwidth Matrices, in my experience, the differences are minimal.

If you like this change, I could adapt the tests so that they pass the results given this new method.

[davenza]

Thank you @carloslihu. I will try to create a test with a case of almost 0 variance to try to control it. First, I wanna know where it is raising the error.

About the diagonal matrices: yes, many authors recommend using diagonal matrices. Less parameters to estimate and usually small degradation of performance. That is why I defined ProductKDE (https://pybnesian.readthedocs.io/en/latest/api/factors.html#pybnesian.ProductKDE).

I know that ProductKDE is not exactly the same as KDE with diagonal bandwidth matrix, but:

• Do you remember if the methods to estimate the bandwidth are equal (in ProductKDE and diagonal KDE)?
• If previous question is no. Do you think we need an specific implementation for the estimation of bandwidth in diagonal KDEs?

[carloslihu]

Hello David,
After analyzing the definition of PyBNesian ProductKDE and the definition of diagonal bandwidth matrix KDE found in the literature.
I think they represent the same concept with different notation.

For my implementation I wanted to use ProductKDE for Hill Climbing, but by lack of skill and time, I wasn't able to do so.
Therefore, my Hack Fix was to manually remove the non-diagonal elements from the covariance matrix in
NormalReferenceRule.hpp and ScottsBandwidth.hpp

This is just a temporary fix, I think that the best solution would be to allow the option of allowing the selection of your ProductKDE implementation for the Hill Climbing algorithm, but my lack of knowledge wouldn't let me.

[carloslihu]

Hello David,
For the moment I have rebased my master branch to tag v0.5.2 (before adapting my diagonal bandwidth fix)
This way you may merge my latest changes without altering your functionality.

I will continue using the diagonal bandwidth in another personal branch (feature/diagonal-bandwidth) for the moment

[Copilot]

🟡 Changes recommended

There are several correctness/build issues in the current diff (test import path, missing C++ includes, an integer-division bug in smoothing, and build configuration pitfalls) that should be addressed before merging.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

This PR modernizes PyBNesian’s Python/C++ codebase via formatting and documentation improvements, while also introducing new learning/independence-test functionality and build/dependency tweaks (including a version bump).

Changes:

• Reformats and refactors many Python tests and updates documentation/changelogs for the 0.6.0 release.
• Updates C++ internals around scoring/operators, KDE bandwidth handling, mutual information behavior, and introduces VP-tree–backed mixed-data mutual information tooling/bindings.
• Adjusts build/packaging configuration (vcpkg, CMake, requirements/pyproject metadata, CI workflows).

File summaries

File	Description
vcpkg.json	Disables Arrow default features in vcpkg manifest.
vcpkg-configuration.json	Reformatting only.
tests/serialization/serialize_models_type_test.py	Import/order formatting and fixture/test formatting cleanup.
tests/serialization/serialize_factor_type_test.py	Refactors tests to use pbn.* types consistently; formatting.
tests/serialization/serialize_factor_test.py	Refactors to pbn.* usage and reformats; minor assertion adjustments.
tests/models/HeterogeneousBN_test.py	Formatting and minor comment cleanup.
tests/models/DynamicBayesianNetwork_test.py	Refactors to shared helpers, reformats, and adjusts regex handling.
tests/models/BayesianNetwork_type_test.py	Refactors to shared helpers and pbn.* usage.
tests/learning/scores/bic_test.py	Uses shared data helpers/constants; formatting.
tests/learning/parameters/mle_test.py	Uses shared data helpers/constants; small assertion additions.
tests/learning/operators/operatorstabuset_test.py	Formatting/whitespace cleanup.
tests/learning/operators/operatorset_test.py	Uses shared data helpers/constants; formatting; variable-name updates.
tests/learning/operators/operators_test.py	Formatting and consistent quoting.
tests/learning/operators/operatorpool_test.py	Uses shared data helpers/constants; adds simple pool creation check.
tests/learning/independence_tests/independence_test.py	New test suite for multiple independence tests.
tests/learning/algorithms/hillclimbing_test.py	Refactors to shared helpers; adds custom BN type/classes in-test; formatting.
tests/learning/algorithms/constraint_test.py	Refactors to pbn.* usage; formatting; adjusts node labels.
tests/helpers/util_test.py	Removes legacy helper module (replaced by tests/helpers/data.py).
tests/helpers/kde.py	Adds helper KDE routines for diagonal bandwidth / logpdf checks.
tests/factors/factor_type_test.py	Refactors to pbn.* base classes and formatting.
tests/factors/discrete/DiscreteFactor_test.py	Uses shared discrete data helper and refactors variable/category conventions.
tests/dataset/holdout_test.py	Uses shared data helpers/constants; formatting.
tests/conftest.py	Alters sys.path handling for test helper imports.
requirements.txt	Removes version specifiers for dependencies.
README.md	Updates badges/headers, repo clone URL, and formatting; links INSTALLATION.md.
pytest.ini	Expands norecursedirs and adds -s default addopts.
pyproject.toml	Version bump to 0.6.0; authors expanded; dependency metadata consolidated.
pybnesian/vptree/vptree.hpp	New VP-tree interface/types for kNN-based algorithms.
pybnesian/util/validate_whitelists.hpp	Adds Doxygen-style documentation blocks.
pybnesian/util/validate_options.hpp	Changes score-validation seed type to unsigned int.
pybnesian/util/validate_options.cpp	Adds documentation/comments; default score selection commentary; minor message text.
pybnesian/util/progress.hpp	Adds formatted_log_t and spinner documentation.
pybnesian/util/basic_eigen_ops.hpp	Refactors PSD check implementation and signature.
pybnesian/pybindings/pybindings_learning/pybindings_independences.cpp	Adds Python bindings for MixedKMutualInformation.
pybnesian/pybindings/pybindings_learning/pybindings_algorithms.cpp	Fixes docstring typo for validated-lik.
pybnesian/opencl/opencl_config.hpp	Adds comments clarifying logsumexp steps.
pybnesian/models/BayesianNetwork.hpp	Adds documentation and comments around blacklist/whitelist handling.
pybnesian/learning/scores/validated_likelihood.hpp	Adds class/method documentation comments.
pybnesian/learning/scores/holdout_likelihood.hpp	Adds class documentation comments.
pybnesian/learning/scores/cv_likelihood.hpp	Adds class documentation comments.
pybnesian/learning/scores/cv_likelihood.cpp	Adds documentation and clarifying comments.
pybnesian/learning/parameters/mle_DiscreteFactor.cpp	Adds smoothing for discrete MLE probability estimation.
pybnesian/learning/operators/operators.hpp	Adds deterministic comparator for delta sorting; adds documentation/comments.
pybnesian/learning/operators/operators.cpp	Adds documentation/comments; refactors scoring update comments; minor logic commentary.
pybnesian/learning/independences/hybrid/mutual_information.cpp	Ensures MI/CMI returns at least machine tolerance; guards pvalue against inf/nan.
pybnesian/learning/independences/hybrid/mixed_knncmi.hpp	New mixed-data kNN-CMI interface declarations.
pybnesian/learning/algorithms/hillclimbing.cpp	Adds extensive function documentation and commentary.
pybnesian/kde/ScottsBandwidth.hpp	Improves messages/comments; enforces diagonal covariance handling and PSD checks.
pybnesian/kde/ProductKDE.hpp	Include order tweak; adds bandwidth-related comments.
pybnesian/kde/opencl_kernels/KDE.cl.src	Expands header comment.
pybnesian/kde/NormalReferenceRule.hpp	Improves messages/comments; diagonal covariance handling adjustments.
pybnesian/kde/KDE.cpp	Adds method documentation comments.
pybnesian/factors/discrete/DiscreteFactor.cpp	Adds clarifying comment in ToString.
pybnesian/factors/continuous/LinearGaussianCPD.hpp	Makes data_type() depend on training type.
pybnesian/factors/continuous/LinearGaussianCPD.cpp	Tracks training type, improves sampling to support float/double, updates pickle state.
pybnesian/factors/continuous/CKDE.hpp	Adds _fit documentation comment.
pybnesian/factors/continuous/CKDE.cpp	Adds fit documentation comment.
pybnesian/dataset/holdout_adaptator.hpp	Fixes formatting in thrown message.
pybnesian/dataset/dataset.hpp	Fixes preprocessor conditional formatting.
pybnesian/dataset/dataset.cpp	Fixes preprocessor conditional formatting; adds documentation comments.
lib/eigen-3.3.7/scripts/relicense.py	Python 3 style/format fixes.
lib/eigen-3.3.7/debug/gdb/printers.py	Python 3 style/format fixes and type-ignore.
INSTALLATION.md	New detailed installation guide.
expand_sources.py	Formatting and minor refactor for file operations.
docs/source/conf.py	Updates version/authors and reformats Sphinx config.
docs/source/changelog.rst	Adds v0.5.2 entry.
docs/source/api/learning/independences.rst	Adds bibliography entries for new MI estimators.
conv_template.py	Formatting, typing, and minor structure cleanup.
CMakeLists.txt	Refactors conditionals; adds ccache; changes python invocation; adds vptree/mixed_knncmi sources.
CHANGELOG.md	Adds entries for 0.6.0/0.5.3 and formatting improvements.
.readthedocs.yaml	Minor formatting fix.
.gitignore	Adds wheelhouse to ignored outputs.
.github/workflows/test_release.yml	Captures/uploads vcpkg logs on failure.
.github/workflows/release.yml	Captures/uploads vcpkg logs on failure.

Review details

• Files reviewed: 86/89 changed files
• Comments generated: 7
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.