Tau Benchmark

Tau tops all rankings regardless of task type and dataset size. Traditional libraries specialize in narrow regimes: generalized linear models and naive Bayes excel on small datasets, while boosting methods dominate on large ones. Tau is the first library to achieve state-of-the-art results across the entire spectrum, from tiny to massive datasets.

The same holds for predictor complexity. Most methods either handle simple main effects well or capture complex interactions, but not both. Tau delivers top performance whether the underlying relationships are additive or highly nonlinear, with no trade-off between the two.

The same applies across industries and subject fields. The benchmark datasets span healthcare, finance, ecology, engineering, social sciences, and more. Tau leads in every domain, showing that its advantage is not tied to any particular data distribution or application area.

Each model's mean rank in competition over each dataset is displayed below. Lower is better.