Enhance counterfactual with fixed_features and custom methods #1029

[mariam851]

Hi @rabst,

Description: This PR introduces enhancements to the create_counterfactual function to address performance bottlenecks and improve flexibility, as discussed in issue #1029.

Key Changes:

Feature Masking (fixed_features): Added a new parameter fixed_features (list of indices) that allows users to specify which features should remain constant during the optimization process. This reduces the search space dimensionality, leading to significantly faster convergence in high-dimensional datasets.

Optimization Methods: Exposed the method parameter for scipy.optimize.minimize, allowing users to choose solvers other than the default 'Nelder-Mead' (e.g., 'BFGS' or 'L-BFGS-B').

Improved Convergence: The internal objective function now dynamically handles the feature mask, ensuring that only the relevant features are updated.

API Consistency: Updated init.py and docstrings to ensure the new parameters are properly exposed and documented.

Note on CI Failures: I noticed that the build-linux and conda unit tests are currently failing. After investigating the logs, the failures are located in mlxtend/classifier/tests/test_perceptron.py with the error: TypeError: only 0-dimensional arrays can be converted to Python scalars.

This seems to be an upstream issue related to recent NumPy updates affecting the Perceptron implementation and is unrelated to the changes in this PR. I have verified that all new and existing tests in mlxtend/evaluate/tests/test_counterfactual.py are passing locally and on the CI.

Fixes: #1029

[rasbt]

Thanks for the PR(s). I have been busy with other projects lately and didn't quite get the chance to check them. I hope to be able to take a look next in the next 1-2 weeks though. I just fixed the perceptron issue, and the CI should work now.

[mariam851]

Hi Sebastian, Thanks a lot for letting me know — really appreciate it. No worries at all, I totally understand how busy things can get. Thanks as well for fixing the perceptron issue, and glad to hear the CI is working now. Looking forward to your feedback whenever you get the time. Best, Mariam في الاثنين، ٥ يناير ٢٠٢٦، ٥:٥٧ ص Sebastian Raschka ***@***.***> كتب:

…

*rasbt* left a comment (rasbt/mlxtend#1148) <#1148 (comment)> Thanks for the PR(s). I have been busy with other projects lately and didn't quite get the chance to check them. I hope to be able to take a look next in the next 1-2 weeks though. I just fixed the perceptron issue, and the CI should work now. — Reply to this email directly, view it on GitHub <#1148 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A5QEUUHAG376OOFAUYTZQL34FHHH3AVCNFSM6AAAAACQHTBFPOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTOMBYG43TCMBWG4> . You are receiving this because you authored the thread.Message ID: ***@***.***>