firthmodels

Summary This release adds optional Numba acceleration for FirthLogisticRegression and FirthCoxPH, greatly accelerating both of them over the numpy/scipy backend. See #12 and #32 for benchmarking results. It also improves numerical robustness when fitting models with rank-deficient or ill-conditioned design matrices, with column-pivoting QR fall-back for logistic regression and pivoted Cholesky for Cox. What's Changed Add optional Numba backend for FirthLogisticRegression by @jzluo in https://github.com/jzluo/firthmodels/pull/35 Add optional Numba backend for FirthCoxPH by @jzluo in https://github.com/jzluo/firthmodels/pull/37 Add QR with column pivoting fallback for compute_logistic_quantities by @jzluo in https://github.com/jzluo/firthmodels/pull/38 Add pivoted Cholesky fallback for compute_cox_quantities by @jzluo in https://github.com/jzluo/firthmodels/pull/39 numba: replace dpotri with dpotrs for matrix inversion by @jzluo in https://github.com/jzluo/firthmodels/pull/40 Full Changelog: https://github.com/jzluo/firthmodels/compare/v0.2.2...v0.3.0