neural network - Counting number of clusters in data using Self Organizing Map -


I know SOM roughly and it is mapping its network to different groups of trained data. How to apply the calculation of the number of groups using SOM? I use the KNL library to apply SOM. In my demo, it is shown how how can I implement this to calculate the number of groups of training and testing? I know I can also use DBSCAN for cluster calculation. But first, I want to implement SOM for cluster calculation. My input data represents the digit in 2D space, such as

  (132.181,0.683431), (136.886,0.988517), (137.316,0.504297), (133.653,0.602269)  

From your question it seems that you think DBSCAN and SOM do the same thing If so, then you are wrong SOM only reorganize your data (and reduce dimension), while DBSKANAN group cluster data. Learning SOM No is a clustering algorithm.

The closest thing to the SO means (which in turn is a clustering algorithm). The only difference between the tools and the SOM is that the SOM (A) is not cluster and b) use the neighborhood function.

(BTW: The difference between Kashmir and DBSCN is that you have to know the number of Kashmir with Kashmir)

Conclusion: Unlike DBSNCN, the SOM does not make any cluster, So you will not be able to count them. If you have a SOM then you will need to run the clustering algorithm on it to get the actual centrosides and clusters.

But why would you do this on earth?
If you need clustering, then choose a clustering algorithm (K-sense / DBSCAN).
If you need dimensional reduction, choose the theme
It seems that whatever you want to achieve, as long as you need groups, you are feeling better keeping the SOM in mind.

Edit:

If for some theoretical reasons, you still want to do this, take a look at this


Comments

Popular posts from this blog

ios - How do I use CFArrayRef in Swift? -

eclipse plugin - Run java code error: Workspace is closed -

c - Error on building source code in VC 6 -