Rolling AUC-PR contribution

[davidlpgomes]

Hey! Last year, my team and I implemented the efficient rolling ROC-AUC on river (pull request).

We would like to contribute again to river by implementing our new efficient algorithm to compute the prequential (rolling) AUC-PR. Our paper was published at the 2023 International Conference on Machine Learning and Applications (ICMLA) and recently made available on IEEE Xplore.

The AUC-PR is a metric related to the AUC-ROC, but was shown to be less optimistic in heavily
unbalanced datasets (Cook and Ramadas, Liu and Bondell), therefore, it is a more suitable metric to evaluate models trained on datasets from certain online machine learning domains, such as credit card fraud and malware classification.

We would like to implement the Rolling AUC-PR in C++, and use Cython to make the code available on Python, just as we implemented the Rolling ROC-AUC, but feel free to make recommendations.

What do you guys think? Let me know your thoughts! 😃