Skip to content

[Bench #478] Make fit/predict/transform/proba separate first-class performance rows #489

Open

Description

@godofecht

Parent: #478

The canonical headline currently classifies an estimator/dataset from fit+predict combined time. That hides phase-local wins and losses: GaussianNB Digits fits faster but predicts slower; RandomForest Digits predicts faster but fits slower; KernelRidge loses in both; KernelSVC has different fit/predict behavior.

Generate machine-readable operation rows for every supported estimator operation: fit, predict, transform, decision_function and predict_proba where applicable. Headline estimator totals may remain as a secondary view, but no total may hide a losing operation.

Acceptance: operation-level median/IQR/speedup/status in frozen artifacts and Pages; operation parity/state prerequisites; each loss links to a focused issue; #478 exit criteria are evaluated over operations, not only estimator totals.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions