Skip to content

0.0.5

Choose a tag to compare

@jaydu1 jaydu1 released this 29 Apr 00:17
· 29 commits to main since this release

[0.0.5] - 2026-04-28

Added

  • Multiple testing correction: conf_int() and summary() now support a multitest_method parameter (mimicking statsmodels) for controlling FDR/FWER. Adjusted p-values are returned as pvalue_adj.
  • Unified Group Importance: conf_int() now supports a groups argument to compute group-level feature importance with uncertainty, replacing the separate group_importance() logic (now deprecated).
    • Accepts groups as a dict of index lists, a 1-D label array, or a binary pandas.DataFrame indicator matrix (features may belong to multiple groups).
    • Optional null-feature thresholding (threshold_null=True) zeros out per-feature UEIFs with negative mean before aggregation.
  • Improved API Consistency: Renamed phi_hat to score in conf_int() output for better clarity.
  • Per-sample UEIFs (ueifs_X, ueifs_Z) are now stored as instance attributes after calling OTExplainer, EOTExplainer, and FlowExplainer, enabling downstream group aggregation.