http://code.activestate.com/recipes/tags/som/2014-06-16T13:20:01-07:00ActiveState Code Recipes2014-06-16T13:20:01-07:00Roberto Bellohttp://code.activestate.com/recipes/users/4189498/http://code.activestate.com/recipes/578893-kindex-for-som-neural-networks-python/

<p style="color: grey"> Python recipe 578893 by <a href="/recipes/users/4189498/">Roberto Bello</a> (<a href="/recipes/tags/cv/">cv</a>, <a href="/recipes/tags/data_mining/">data_mining</a>, <a href="/recipes/tags/means/">means</a>, <a href="/recipes/tags/neural_network/">neural_network</a>, <a href="/recipes/tags/python/">python</a>, <a href="/recipes/tags/som/">som</a>). </p> <p>Achieved cataloging into groups by a SOM neural network, the question arises whether or not there is knowledge in the groups, namely whether the groups are between them distinct and have homogeneous characteristics within each group. The use of the coefficient of variation (CV) can be of help.</p> <p>KINDEX (Knowledge Index) is an index that measures how much knowledge is contained in the groups obtained from the SOM neural network: in the case KINDEX reaches the maximum value of 1, each group would consist of records with constant values ​​in all the variables / columns, and each group would be quite distinct from other groups. KINDEX is calculated using the weighted-average CV of variables / columns groups, comparing them to the CV of the variables / columns of the input file before cataloging.</p>