PL-kNN: A Python-based implementation of a parameterless k-Nearest Neighbors classifier [Formula presented]

Abstract

This paper presents an open-source implementation of PL-kNN, a parameterless version of the k-Nearest Neighbors algorithm. The proposed model, developed in Python 3.6, was designed to avoid the choice of the k parameter required by the standard k-Nearest Neighbors technique. Essentially, the model computes the number of nearest neighbors of a target sample using the data distribution of the training set. The source code provides functions resembling the Scikit-learn methods for fitting the model and predicting the classes of the new samples. The source code is available in the GitHub repository with instructions for installation and examples for usage.