vfi: Classification by Voting Feature Intervals in Python

vfi is a Python package that mainly implements a variant of a deterministic supervised learning method that is named classification by Voting Feature Intervals (hereinafter VFI). VFI was proposed by Demiröz and Güvenir in order to cope with supervised classification problems and until now it was only available in Weka platform. The proposed approach works similarly to Naive Bayes, in the sense that it works independently on each feature. Constructs intervals for each feature and class counts are recorded for each interval for each feature. The classification of an unseen instance is performed using the features' interval confidence for each class. The vfi package provides a drop-in replacement class for scikit-learn compatible estimators and depends on scikit-learn and numpy. The proposed implementation differs from the original implementation in two folds. Firstly, it gives the ability to the user to define the number of the bins for the discretization of each feature, as well as the automatic method that is proposed in the original paper. Secondly, it gives the ability to the user to select the strategy of the discretization process in order to create the intervals for each feature. The available strategies are: uniform where the bins in each feature have identical widths, quantile where the bins in each feature have the same number of points and kmeans where the values in each bin have the same nearest center of a 1d k-means cluster. The last two aforementioned strategies are in addition to the original implementation, which is available in Weka, where the strategy that is used for the construction of the intervals is uniform. VFI is simple to understand, has over 200 citations in the scientific literature, is a surprisingly fast algorithm and performs very well in most of the cases. The source code is available at https://github.com/chkoar/vfi while the documentation is rendered at https://vfi.readthedocs.io.