load_tdc_splits

skfp.datasets.tdc.load_tdc_splits(dataset_name: str, data_dir: str | PathLike | None = None, as_dict: bool = False, verbose: bool = False) → tuple[list[int], list[int], list[int]] | dict[str, list[int]]

Load pre-generated dataset splits from the TDC benchmark.

TDC [1] uses precomputed scaffold split with 80/10/10% split between train/valid/test subsets. Those splits are widely used in literature and allow for a realistic estimate of model performance on new data.

Dataset names here are the same as returned by load_tdc_benchmark function, and are case-sensitive.

Parameters:

• dataset_name (str) – Name of the dataset to loads splits for.

• data_dir ({None, str, path-like}, default=None) – Path to the root data directory. If None, currently set scikit-learn directory is used, by default $HOME/scikit_learn_data.

• as_dict (bool, default=False) – If True, returns the splits as dictionary with keys “train”, “valid” and “test”, and index lists as values. Otherwise, returns three lists with splits indexes.

• verbose (bool, default=False) – If True, progress bar will be shown for downloading or loading files.

Returns:

data – Depending on the as_dict argument, one of: - three lists of integer indexes - dictionary with “train”, “valid” and “test” keys, and values as lists with splits indexes

Return type:

tuple(list[int], list[int], list[int]) or dict

References

[1]

Huang, Kexin, et al. “Therapeutics Data Commons: Machine Learning Datasets and Tasks for Drug Discovery and Development” Proceedings of Neural Information Processing Systems, NeurIPS Datasets and Benchmarks, 2021