lucie: An Improved Python Package for Loading Datasets from the UCI Machine Learning Repository.

Area of Science:

• Machine Learning
• Data Science
• Computer Science

Background:

• The University of California-Irvine (UCI) Machine Learning Repository (UCIMLR) is a vital resource for high-impact datasets.
• A significant portion of UCIMLR datasets, particularly those in .zip files with nonstandard formats, are difficult to import using the recommended ucimlrepo package.

Purpose of the Study:

• To develop a utility that automates the import of previously non-importable datasets from the UCIMLR.
• To preserve the tabular data structure of these datasets during import.

Main Methods:

• A new Python package, lucie (load University California Irvine examples), was developed.
• lucie automatically determines data formats and facilitates import.
• The utility was designed using the top 100 UCIMLR datasets and benchmarked on the subsequent 130.

Main Results:

• lucie achieved a 95.4% success rate in importing challenging datasets, compared to 73.1% for ucimlrepo.
• The package demonstrates high performance on datasets previously inaccessible via automated methods.
• lucie offers 98% code coverage and is available on PyPI.

Conclusions:

• lucie significantly enhances accessibility to a large number of UCIMLR datasets.
• The utility addresses a critical gap in data import for machine learning research.
• lucie provides a robust and efficient solution for data scientists and researchers.