BSOM Train a SOM using the batch SOM training algorithm
CALLS
[som,mse] = bsom(p)
[som,mse] = bsom(p,smap)
[som,mse] = bsom(p,smap,epochs,Nc)
[som,mse] = bsom(p,smap,epochs,Nc,weights)
[som,mse] = bsom(p,smap,epochs,Nc,weights,Bidx)
INPUTS
p: Input data matrix (R x Q) where each column correspond to each sample vector
in the input space.
smap: SOM structure (see INISOM)
epochs: Number of times (iterations) which the whole input data set is used for
training
Nc: Neighbourhood. It can be either one of these:
- a single scalar specifying the last value of a monotonically decreasing
neighbourhood.
- a (1 x 2) vector with the starting and finishing value of the neighbourhood
- a (1 x epochs) vector with values of the neighbourhood at each epoch
weights: Weight values given to all the samples (by default all samples have the same
weight)
Bidx: Index vector containing variables to be taken into account in the
self-organization process
EXAMPLE
Trains a SOM ordered by variables 2 and 3
p = rands(3,500);
p = [p(1:3,:); [(1-p(1,:).^2)].*[(1-p(2,:).^2)].*[tansig(p(3,:))]; randn(2,500)];
som = inisom(p,{30,30});
som = bsom(p,som,10,1,ones(1,size(p,2)),[2,3]);
planes(som)
OUTPUTS
som: Self-Organizing Map structure
mse: Vector with the mean squared quantization error for each epoch.
DESCRIPTION
Trains the SOM parmeters specified in the smap structure using the so-called batch SOM
algorithm.
By default uses a monotonically (logaritmic) decreasing neighbourhood.
It also allows to specify weight values to account for the contribution of each
sample to the density function in the input space.
See also INISOM, PLANES, PLANES3D, SOMINTERP, FPROJ, BPROJ