parallelpam (Q82394)

Application of the Partitioning-Around-Medoids (PAM) clustering algorithm described in Schubert, E. and Rousseeuw, P.J.: "Fast and eager k-medoids clustering: O(k) runtime improvement of the PAM, CLARA, and CLARANS algorithms." Information Systems, vol. 101, p. 101804, (2021). <doi:10.1016/j.is.2021.101804>.It uses a binary format for storing and retrieval of matrices developed for the 'jmatrix' package but the functionality of 'jmatrix'is included here, so you do not need to install it. Also, it is used by package 'scellpam', so if you have installed it, you do not needto install this package.PAM can be applied to sets of data whose dissimilarity matrix can be very big. It has been tested with up to 100.000 points.It does this with the help of the code developed for other package, 'jmatrix', which allows the matrix not to be loaded in 'R' memory (whichwould force it to be of double type) but it gets from disk, which allows using float (or even smaller data types). Moreover, thedissimilarity matrix is calculated in parallel if the computer has several cores so it can open many threads. The initial partof the PAM algorithm can be done with the BUILD or LAB algorithms; the BUILD algorithm has been implemented in parallel. The optimizationphase implements the FastPAM1 algorithm, also in parallel. Finally, calculation of silhouette is available and also implemented in parallel.