Changelog

Version 2.0.0

  • Added binary logistic regression (combss.logistic).

  • Added multinomial logistic regression (combss.multinomial).

  • Added Frank-Wolfe homotopy algorithm as the new default method for linear regression (combss.linear.model.fit(method='fw')).

  • Added LOOCV-based ridge penalty selection (combss.cv).

  • Original Adam + dynamic-lambda method remains available via combss.linear.model.fit(method='original').

  • Added scikit-learn and pandas as dependencies.

Version 1.1.4

  • Linear regression via Adam optimiser with dynamic lambda grid.

  • Performance metrics for variable selection evaluation.