Code for SPOCC, a Scalable POssibilistic Combination of Classifiers

An ensemble method that relies on a possibility theory and confusion matrices arXiv.

The repository contains four python files.

It can be readily downloaded and executed in a pyhton console (python 3) provided that the imported python module versions on your machine match the following ones.

To have a quick look on the performances of SPOOC run the example() function in spocc_main.py The default parameter values allows to obtain a quick comparison between SPOOC and reference methods (classifier selection, weighted vote aggregation, stacking, exponential weights, naive Bayes and Bayes aggregation or centralized learning) on simple synthetic datasets. To achieve a prescribed confidence level in the returned accuracies the parameter iter_max must be set to np.inf but the execution will be significantly longer.

Warning: the code is compatible with these module versions: numpy 1.17.2, matplotlib 3.1.1, sklearn 0.22.1, scipy 1.4.1

This software is distributed under the CeCILL Free Software Licence Agreement