Hi everyone, I just came across this site. Apologies if this is the wrong forum to post this question in.

I'm having a problem with self organising maps. I understand the theory and algorithms but I'm not sure how to go about implementing it in code (C#). I've seen a couple of examples on http://www.ai-junkie.com/ann/som/som1.html and http://davis.wpi.edu/~matt/courses/soms/#Java
but neither are C# (which is the only language I'm familiar with). This isn't a problem algorithmically as that's easy to convert, but I'm having great difficulty on the user interface side. What's the best way of testing my SOM code? I don't know where to start with the interface side. There's no need for code (unless you have C# code) but theory as to what's actually happening in the previous sites and the how I should be rendering the nodes etc..
Thanks to anyone who can shed any light on this. I've been trying to figure it out for weeks, and I just can't get started :(

A common way to test data mining predictive models (backpropagation neural nets, CART etc etc) is to validate the model against unseen (control group) outcome data where the outcome (or member classification) is known.

What type of output would your model produce?  There's not much point making an SOM unless you can use it to score new data (unless the SOM is some fancy super scalable clustering and index 100% of everything akin to google methodolgy...)

Commercial tools such as Clementine (which i use) displays a simple Kohonen 2D non-cyclic SOM in a square grid.  It also scores new data with two columns.  One column indicates the y-axis of the SOM and the other column indicates the x-axis, so that together you know which cluster any row is allocated to.

Hope this helps

Cheers

Tim