Similarity coloring of a self organizing map

I have implemented the algorithm to train self organizing maps in Python and it seems to be working well. I checked with some labeled data and the maps are learning the topology well. Here are some images:

I am calculating a Unified Distance Matrix for visualization which does not seem to be working well as seen in the 2nd image.

I believe that similarity coloring as seen in the world poverty map developed by Prof. Kaski and Prof. Kohonen would be apt for my application but I am unable to find a clear explanation of how it is to be calculated. The procedure has been outlined in the paper "Tips for Processing and Color-coding of Self-Organizing Maps ", however, it omits some technical details which are necessary to implement the similarity coloring. Here is the image that I am referring to:

Could someone help me with the exact steps to calculate the similarity coloring map? I have searched everywhere for a month and I am unable to find an explanation.

Topic neural-network

Category Data Science

1
answered at

I don't know exactly how the above was created, but I did something similar in the late '90s and reproduced it recently using somoclu in Python. While I've not applied the colours back to the SOM cells, this would be pretty trivial.

I've put the notebook on kyso, I hope it's useful.

About

Geeks Mental is a community that publishes articles and tutorials about Web, Android, Data Science, new techniques and Linux security.