load_duo2018

datasets.load_duo2018()

Load the bundled Duo 2018 clustering benchmark.

Reads DuoSCClustering2018.csv from the installed package via importlib.resources so it works whether beam runs from a source checkout or an installed wheel. The wide CSV has one method per row and one column per metric and data set combination, named <prefix>_<dataset>. The function reshapes it into a (n_methods, n_datasets, n_metrics) tensor with NaN where the source held the literal string NA.

Returns

Type	Description
Duo2018	Frozen dataclass with method names, data set names, metric ids, per-metric polarity, and the (14, 12, 4) score tensor.

Notes

No imputation or dropping happens here. The missing cells (5 each in ari, runtime and shannon_entropy_diff, 101 in nclust_deviation) are left as NaN. Use the coverage helpers to decide on a policy.