SOM tutorial with a 2D test set

Imports



In [1]:

    
import sys
sys.path.append('.')
sys.path.append('../')
import generatedatasets
import SOM2
import SOMclust
from pylab import *
pylab.rcParams['figure.figsize'] = 10, 10  # that's default image size for this interactive session

Generate dataset



In [2]:

    
xy, c = generatedatasets.dataset_fixed_cov()



In [3]:

    
scatter(*xy.T, c=c)









    Out[3]:





<matplotlib.collections.PathCollection at 0x1015ce90>



In [4]:

    
xy.shape









    Out[4]:





(4000, 2)

Learn



In [5]:

    
som = SOM2.SOM(xy)



In [6]:

    
smap = som.learn(verbose=True)









    



1 7999 8000 99.99% 1.00013637015 1.13641787895e-05 (25, 34)

and U-matrix



In [7]:

    
umat = som.umatrix()



In [8]:

    
matshow(umat)
colorbar()









    Out[8]:





<matplotlib.colorbar.Colorbar instance at 0x2ba7f9260830>

Compute BMUs -- the position of each input data in the SOM



In [9]:

    
bmus = som.get_allbmus()

Flood the periodic map



In [10]:

    
clust = SOMclust.clusters(umat, bmus, som.smap)









    



100.00/100: flooding: 2499/2499, 1.28, (145, 160)



In [11]:

    
imshow(ma.masked_array(clust.umat_cont, clust.mask), interpolation='nearest')









    Out[11]:





<matplotlib.image.AxesImage at 0x2ba7f9527210>