Quickstart

The easiest way to run the benchmark is to to use a pre-trained model. The models files can be downloaded here and should be placed data/models folder

    from tabularbench.benchmark.benchmark import benchmark
    clean_acc, robust_acc = benchmark(
        dataset="URL",
        model="STG_Default",
        distance="L2",
        constraints=True,
    )
    print(f"Clean accuracy: {clean_acc}")
    print(f"Robust accuracy: {robust_acc}")