SOMDisco is an R package intended to address several advances in SOM learning and analysis that are currently missing in currently available SOM packages for R (som, kohonen, and popsom). The main contributions of SOMDisco are:
- Integration of DeSieno's Conscience-SOM modifications to Kohonen's original algorithm to attempt maximum entropy SOM learning
- Fast and efficient C++ implementation of CSOM training (based on Rcpp and RcppArmadillo)
- Optional parallel SOM training and recall (as applicable, via RcppParallel)
- Computation of the CADJ matrix, which is a weighted topological adjacency of SOM prototypes helpful in cluster discovery
- The CONNvis visualization (requires the
TopoRNet
R package) which represents the topological information in CADJ on the SOM lattice - Other advanced SOM visualizations (the mU-matrix, propagation of discrete and continuous values on the lattice)
devtools::install_github("somdisco/SOMDisco")
See the SOMDisco homepage for more information.