About Self Organizing Maps

Hi, the question is about training number of Self Organizing Maps(SOM) function in Matlab that we need to minimize the error between the samples and the Best Matching Units (BMU). In neural network toolbox we can observe the U-matrix and component planes. Maybe, when the patterns of U-matrix or component planes are close to stable, we can stop training. Is this the solution? Are there any parameters to decide the iteration number? For example, in SOM Toolbox (Laboratory of Computer and Information Science) for Matlab 5, there are some functions that calculate quantization error (average distance between each data vector and its BMU)and topographic error (the proportion of all data vectors for which first and second BMUs are not adjacent units) Is there any function to calculate that kind of errors and SOM quality in Matlab? Many thanks...

[Peter]

Did you find a simple solution to this problem. I've reached the same point as well. Thanks

[Dorothee Hohensee]

It's now 4 years ago... Did you find any solution? That would be very helpful for me!

Answer 1

[Shahrbanoo Hazratiyadkoori]

Hi, I have problem with quantization of SOM maps in MATLAB, also if you know how I could calculate average quantization error and topographic error, please help me in these issues. Thanks

[Bryn Ronalds]

How are there no responses to this question? I am also asking the same thing..

Given that the SOM is unsupervised there is no performance outputs within the training record, and certainly none output iteratively at every epoch so as to track the minimization of the error (default appears to be "mse").

Any advice on how to either calculate this manually, or ideally have it be output within the network, would be much appreciated, thanks.