3
$\begingroup$

I have a large dataset I am trying to do cluster analysis using SOM. The dataset is huge (~ billions of records) and I am not sure what the number of neurons should be or the SOM grid size to start with.

Any pointers to some material that talks about estimating the number of neurons and grid size would be greatly appreciated.

Improve this question
$\endgroup$

2 Answers 2

Reset to default
1
$\begingroup$

Kohonen's Self Organizing Maps (1995) says that the SOM is an approximation of some density function, p(x) and the dimensions for the array should correspond to this distribution. "Therefore visual inspection of the rough form of p(x), e.g. by Sammon's mapping ought to be done first." p.112

Improve this answer
$\endgroup$
0
$\begingroup$

This is my heuristics:

Look at eigenvalues resulted from SVD and watch for relatively large values. Use number of units as the number of higher eigenvalues.

For grid size, take the ratio of largest eigenvalue to second and use that ratio to decide proportion of the map heigh and width

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.