[antonsoo]

Fixes #703.

CustomLibrary stores its functions as functions, but scikit-learn looks up the constructor parameter library_functions. This breaks notebook display, clone() and set_params() on both 2.1.0 and current main:

import numpy as np
from sklearn.base import clone
from pysindy import CustomLibrary

library = CustomLibrary([lambda x: x])
print(library.fit(np.ones((10, 2))))
clone(library)
library.set_params(library_functions=[lambda x: x**2])

Expose library_functions as a read/write alias of the existing functions attribute. This also makes parameter replacement affect the functions used by fit and transform, while retaining the documented attribute and libraries saved by older versions.

Fitted libraries cache generated function_names. When changing the number of functions after a fit, also set function_names=None to regenerate names on the next fit.

Checked fitted/unfitted cloning, parameter replacement, an old-writer serialized library, and cloning a fitted Lorenz model. Existing feature arrays, coefficients and predictions remain identical. A scikit-learn Pipeline/GridSearchCV on a recorded waveform now runs; both candidate scores agree with independently assembled matrices and NumPy least squares within 2e-16. This is an API check, not a physiological model.

Validation: all three new regressions fail on base and pass here; feature-library tests: 93 passed, 1 skipped; configured pre-commit checks pass. The complete suite has the same 35 Gurobi max_iter failures on both revisions (base 709 passed, head 712 passed, both 13 skipped and 11 deselected).

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-01T21:35:54.498592Z	39c1560	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.