SOMbrero
, which means 'Self Organizing Maps Bound to Realize Euclidean
and Relational Outputs', implements several variants of the stochastic
Self-Organising Maps and is able to handle numeric and non numeric data sets.
See help(SOMbrero)
for further details.
The numeric SOM is illustrated on the well-known iris
data set. This data
describe iris flowers with 4 numeric variables
(Sepal.Length
, Sepal.Width
, Petal.Length
and Petal.Width
) and a fifth variable (not used to train the SOM) is the
flower's species. This example is treated in the
numeric SOM guide.
The SOM algorithm provided by the package SOMbrero
can also handle some
non-numeric data. First, data described by contingency tables, which can be
processed using the 'korresp' algorithm (see Cottrell et al., 1993). This case
is illustrated on the presidentielles2002
data set which contains, for
each of the French administrative departments (row variables) and each of the
candidates (column variables), the number of votes in the first round of the
2002 presidential election. This example can be found in the
korresp user guide.
Data described by a dissimilarity matrix can also be processed by SOMbrero
as described in Olteanu et al., 2012. This case is illustrated on a data set
extracted from the novel Les Miserables
, written by the French author
Victor Hugo and published in the 19th century. This data set provides a
dissimilarity matrix between the characters of the novel, based on the length
of shortest paths in a network defined from the novel. This example is provided
in the relational user guide.
For those who have an R developer soul, and who want to help improve this package, the following picture provides an overview the current arborescence of the package: