CLUSTERING THROUGH OPTIMAL BAYESIAN CLASSIFICATION

The package contains function for performing soft clustering.

A new soft clustering algorithm is presented (Clustering through Optimal Bayesian Classification). The algorithm does not depend on random initializations, and it contains a native metric to select the optimal number of clusters.

The clustering algorithm minimizes the log-Bayesian risk (classification error) which can be expressed from the soft cluster assignment. The log-Bayesian risk is given as an optimization functional to perform the clustering. An Expectation-Maximization-like algorithm is proposed in order to minimize the proposed functional.

CPU and GPU implementations are included.