pyml-datasets#

Classic tabular datasets for pyml, stored as NumPy .npz files and loaded with a single function call. Only datasets with an explicit CC0 1.0 or CC BY 4.0 license are bundled, and each one keeps its original license, citation and a list of the changes made to it.

This is a personal project of a student, made for pyml. It is not looking for contributors, see the README.

The data is read with NumPy alone. The command line also uses rich and rich-argparse, which belong to the optional cli extra (see Installation) and are not loaded by import pyml_datasets. The tools used to download and convert the original data, for example scikit-learn and ucimlrepo, are needed only to rebuild the data files, never to use them.

Each dataset is described in the Datasets section, with its source, citation, and license. A file is read from disk only when its load_* function is called, so importing the package stays cheap.

Originally built to support the pyml project, but it is a standalone package with no dependency on pyml.

Quick start

Load a dataset, inspect it and train a pyml model in a few lines.

Datasets

What each dataset contains, where it comes from and how to cite it.

License and citation

Which license applies to the code and to each dataset.

API reference

The Dataset class and the load_* functions.