A central task in machine learning and data science is the comparison and selection of models. The evaluation of a single model is very simple, and can be carried out in a reproducible fashion using the standard scikit pipeline. Organizing the evaluation of a large number of models is tricky; while there are no real theory problems present, the logistics and coordination can be tedious. Evaluating a continuously growing zoo of models is thus an even more painful task. Unfortunately, this last case is also quite common.

Reskit is a Python library that helps researchers manage this problem. Specifically, it automates the process of choosing the best pipeline, i.e. choosing the best set of data transformations and classifiers/regressors. The core of reskit is two classes: Pipeliner and Transformer.

First and second sections describe work of this classes. The third section explains how to use this classes for machine learning on graphs.

You can view all tutorials in format of jupyter notebooks here.