pivmet: an R package proposing pivotal methods for consensus clustering and mixture modelling

We introduce the R package pivmet, a software that performs different pivotal methods for identifying, extracting, and using the so-called pivotal units that are chosen from a partition of data points to represent the groups to which they belong. Such units turn out be very useful in both unsupervised and supervised learning frameworks such as clustering, classification and mixture modelling. More specifically, applications of pivotal methods include, among the others: a Markov-Chain Monte Carlo (MCMC) relabelling procedure to deal with the well-known label-switching problem occurring during Bayesian estimation of mixture models (Egidi et al., 2018; Frühwirth-Schnatter, 2001; Richardson & Green, 1997; Stephens, 2000); model-based clustering through sparse finite mixture models (SFMM) (Frühwirth-Schnatter & Malsiner-Walli, 2019; Malsiner-Walli et al., 2016); consensus clustering (Strehl & Ghosh, 2002), which may allow to improve classical clustering techniques—e.g. the classical 𝑘-means—via a careful seeding; and Dirichlet process mixture models (DPMM) in Bayesian nonparametrics (Escobar & West, 1995; Ferguson, 1973; Neal, 2000).