optimal-centroids

Source code for paper Search-based framework for transparent non-overlapping ensemble models by B.Gulowaty and M.Woźniak.

Building the package

Project uses poetry as build system. To install the package for usage in your local env, simply issue poetry install.

Usage

from optimalcentroids.optimal_centroids import run

# define variables
models = run(x_train, y_train, n_trees, max_tree_depth, pop_size, n_gen)

• Method run returns list of models. If, during the optimization process, pareto solutions were found, then the list contains all models based on the pareto front solutions. Otherwise, list contains single model.
• Returned models are compatible with scikit-learn API.

Please refer to this notebook for more extensive usage example.

Parallelization

The inner working are based on pymoo optimization library. Method run accepts pymoo_elementwise_runner keyword argument, which refers to elementwise evaluation function. Default one is LoopedElementwiseEvaluation.