Skip to content

4.3.2

Latest

Choose a tag to compare

@YikaiZhang95 YikaiZhang95 released this 02 Aug 21:09

What's new in 4.3.2

Fixes #6 — solvers no longer fail silently when max_iter is too low.

  • cvksvm, cvkdwd, and cvklogit now track per-lambda convergence and raise an
    aggregated ConvergenceWarning (sklearn.exceptions.ConvergenceWarning, re-exported
    as torchkm.ConvergenceWarning) when the maxit iteration cap is hit without meeting
    the convergence/KKT tolerance, plus a separate warning when the cross-validation path
    exhausts its nlam * maxit budget.
  • The warnings propagate through TorchKMSVC.fit(), TorchKMDWD.fit(), and
    TorchKMLogit.fit(), which also expose the per-lambda status as the new fitted
    attribute converged_.

Full Changelog: 4.3.1...4.3.2