Lantern Court

Compare models with clarity.

Side-by-side metrics, thoughtful caveats, and statistical context for honest decisions.

How it works Learn about metrics

A row of stone lanterns receding into mist, the nearest one lit

Live on

eight real models, ranked live in your browser

Kernel asleep

Lantern Court

Rank	Model	Accuracy	Macro F1	Time

ranked by accuracy · 5-fold CVcomputed in your browser via Pyodide

Fair by construction

The leaderboard on the left is computed live on the dataset you picked. Every model sees the identical 5-fold cross-validation, so the comparison measures the model, not luck in the split. The winner is lit in gold.

Same folds

Every model is scored on the exact same StratifiedKFold splits.

Honest metrics

Accuracy and macro-F1 are averaged across folds, not a single lucky run.

Is the gain real?

A McNemar test tells you whether the top model's edge is signal or noise.

Reproduce

import breezeml
df = breezeml.datasets.wine()
breezeml.compare(df, "wine_class")

Learn about the metrics

Next: cross the Honest Gate. A high score means nothing until it survives scrutiny.

Continue to the Gate