HUGIML PMLBmini Benchmark Dashboard

Official OpenML train/test indices, fold-local model selection, predictive performance, and model-inspection evidence.

Overall model comparison

Performance, ranks, inspection efficiency, and runtime

Performance, ranks & runtime

HUGIML inspection efficiency — baseline / HUGIML

Ratios are baseline model inspection units divided by each HUGIML variant?s model inspection units, on tasks where that variant?s AUC is within tolerance.

Model inspection units versus AUC

One point per completed task/model pair

The x-axis is logarithmic when every displayed inspection value is positive.

Dataset explorer

Inspect paired results for one OpenML task

Dataset-level metric comparison

Three measures × all models

Each row compares ROC-AUC, balanced accuracy, and F1 for the same OpenML task. Bold marks the best score; italics mark the second-best score. Ranking is based on the displayed four-decimal values, so displayed ties receive identical styling.

HUGIML fitted paths and RPTE distributions

Actual route counts and official-split structural distributions

Inputs are the downstream columns supplied to each binary RPTE estimator. Trees and leaves are totals in the fitted one-vs-rest model; path length is averaged across generated terminal leaves.

HUGIML variant paired tests

Wilcoxon tests with Holm adjustment

Evaluation protocol

Run configuration
Methodology and parameter search space

Cross-validation protocol

    Preprocessing and timing

      Complexity unit definitions

        Model search spaces