Benchmark Results

Comparison of counterfactual methods across datasets.

Summary

Method	Validity	Proximity	Sparsity	Plausibility	Speed
PPCEF	0.95	0.82	0.71	0.89	Medium
DICE	0.98	0.75	0.65	0.62	Fast
DiCoFlex	0.94	0.80	0.73	0.85	Medium
GLOBE-CE	0.91	0.88	0.78	0.71	Fast
ReViCE	0.93	0.84	0.74	0.82	Medium

Per-Dataset Results

Adult Dataset

Method	Validity	Proximity (L2)	Sparsity
PPCEF	0.96	0.45	3.2
DICE	0.99	0.52	4.1
GLOBE-CE	0.92	0.38	2.8

COMPAS Dataset

Method	Validity	Proximity (L2)	Sparsity
PPCEF	0.94	0.41	2.9
DICE	0.97	0.48	3.5
GLOBE-CE	0.89	0.35	2.4

Runtime Comparison

Method	CPU (s/instance)	GPU (s/instance)
PPCEF	0.15	0.02
DICE	0.08	N/A
GLOBE-CE	0.05	0.01

Reproducing Results

See Running Benchmarks for instructions on reproducing these results.