Splittind And Group Recombining Algorithm

This is the implementation of the SAGRA algorithm, proposed by Álvarez and Peña (2014)

SAGRA is a new clustering methodology which, based in a strategy of splitting, cleaning and recombining, is able to detect groups inside a data sample. As splitting rule we use the discriminator function, where points sharing the same discriminator are classified into the same group, defining in this way a partition of the sample. For cleaning we detect and purge the outliers of each group, and finally for recombining, we propose the use of a Bayes factor to weight the likelihood of the sample given two models: one where all data is generated from a single distribution, and other when the distribution is a mixture estimated from the obtained partition.

The main function of the package is:

• sagra() takes a data frame and return a clustering solution.

This is a development version, mainly intended for internal improvements since many bugs are expected. Nevertheles, this version can be installed using:

# install.packages("devtools")
devtools::install_github("adolfoalvarez/SAGRA")