Skip to contents

matsketch 0.1.0

  • First release.
  • rpchol(): randomly pivoted Cholesky, in its simple and accelerated (rejection-sampling) forms, with greedy and uniform pivoting as baselines.
  • trace_est() and diag_est(): the XTrace, XNysTrace and XDiag estimators, with Hutch++ and Girard-Hutchinson for comparison.
  • nystrom(), nystrom_precond(), effective_dim() and pcg(): randomized Nystrom preconditioning for conjugate gradients, solving many right-hand sides at once.
  • reml_sketch() and reml_exact(): variance components by average-information REML, with the covariance matrix never formed in the sketched fit.
  • kernel_matrix() and grm_matrix(): kernel and genomic relationship matrices evaluated only where needed.