Abstract
Statistical tests has arisen as a reliable procedure for the validation of results in many kind of problems. In particular, due to their robustness and applicability, non-parametric tests are a common and useful tool in the process of design and evaluation of a machine learning algorithm or in the context of an optimization problem. New trends in the field of statistical comparison applied to the field of algorithms' performance comparison indicate that Bayesian tests, which provides a distribution over the parameter of interest, are a promising approach.
In this contribution rNPBST (R Non-Parametric and Bayesian Statistical tests), an R package that contains a set of non-parametric and Bayesian tests for different purposes as randomness tests, goodness of fit tests or two-sample and multiple-sample analysis is presented. This package constitutes also a solution which integrates many of non-parametric and Bayesian tests in a single repository.