SOMbrero implements the stochastic (also called on-line) Self-Organising Map (SOM) algorithm for numeric and relational data. Data are clustered into a 2-dimensional squared grid that can be initialized with initGrid. The main function is trainSOM that implements three types of algorithms:
-
the standard numeric SOM as in [Kohonen, 2001]
-
the relational SOM designed to deal with datasets described by a dissimilarity as in [Olteanu & Villa-Vialaneix, 2015] and [Mariette et al., 2017]
-
the KORRESP method that deals with contingency tables as in [Cottrell et al, 2004] and [Cottrell et al, 2005]
Results can be displayed with the function plot.somRes and quality criteria are provided by quality. Finally, a super-clustering can be computed with superClass.
Kohonen T. (2001) Self-Organizing Maps. Berlin/Heidelberg: Springer-Verlag, 3rd edition.
Cottrell M., Ibbou S., Letremy P. (2004) SOM-based algorithms for qualitative variables. Neural Networks, 17, 1149-1167.
Cottrell M., Letremy P. (2005) How to use the Kohonen algorithm to simultaneously analyse individuals in a survey. Neurocomputing, 21, 119-138.
Mariette J., Rossi F., Olteanu M., Mariette J. (2017) Accelerating stochastic kernel SOM. In: M. Verleysen, XXVth European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN 2017), i6doc, Bruges, Belgium, 269-274.
Olteanu M., Villa-Vialaneix N. (2015a) On-line relational and multiple relational SOM. Neurocomputing, 147, 15-30.