The Shapley Value
Estimator Benchmark

—estimators

—game settings

—evaluations

Loading…

Leaderboard

Elo comparison panel Estimator variants

Plot family Methods to chart

Lower is better ↘

Median nMSE · ≥80% coverage

Time measurement

Family-balanced median · Comparable timing profiles

Summary metric

By first publication · ≥80% coverage.

Publication sources

Details for these results ＋

Elo.

Hardware.

Get the frozen games and benchmark results ↗

Under-budget runs ＋

Failed runs · top 6 ＋

Download selection

Target Budget

Ridge regression · full benchmark

How the coefficients are fitted

Ridge penalizes large weights. Test datasets enter neither fitting nor tuning. Weights describe combinations, not importance.

Nearby methods have similar projected errors.