"may i ask a question about clustering by som"

joe

Dear Sir
May i ask  a question about clustering by  self-organizing map ?
Does self-organizing map cluster  only through visual map?
Does  self-organizing map have another style(type of result) to cluster? ???
I am so sorry to disturb.
Thanks a lot

MariusHelf

Hi Joe,

can you explain more detailed what you are trying to do? Which operators are you using?

Best regards,
Marius

joe

Dear sir
I want to  cluster by using SOM.
Does self-organizing map cluster  only through visual map?
But i don't know self-organizing map have another style(type of result) to cluster?
I don't want to Determine the number of clusters artificially.
I want to  determine the number of clusters by using som and give the results of SOM to the other operations automatically .
In other words, how to  import the parameters of SOM to the next operation in the workflow?
thanks a lot    
http://photo.xuite.net/chinchaowei/5338584/3.jpg

MariusHelf

Hi Joe,
not determining the number of clusters artificially is a bit difficult with the som, since you specify the number of "clusters" implicitly with the net size parameters: each cell of the som's grid corresponds to one cluster. If you want to know in how many grid cells the som actually placed at least one datapoint, you can use the Remove Duplicates operator (with "include special attributes" disabled) after applying the SOM operator and then count the examples. You can extract the example count with the help of the Extract Macro operator.
Btw, you can also use the SOM directly for clustering by creating a SOM with dimensionality 1.

Best regards,
Marius

joe

thank your help
i try to creat your suggestion.
thanks a lot

joe

dear sir
i try it
but i have I have some problems(doubts)
i don't know how to set the Parameters?(in order to give the clusters of SOM to Flatten Clustering automatically )
may you help me,please?
i will submit Data reports  to  Manager
thank a lot :-*
http://photo.xuite.net/chinchaowei/5338584/2.jpg
http://photo.xuite.net/chinchaowei/5338584/1.jpg

MariusHelf

Hey Joe,

I don't know what your task is, but if you want to cluster by SOM you basically have two options:
1. apply the SOM operator to create a 1-dimensional SOM. Then you will end up with one integer attribute, whose values can be interpreted as clusters.
2. apply the SOM operator to create a multi-dimensional SOM. Then you get a discrete, multi-dimensional space, where on each point zero or more datapoints may lay. You may want to use something like k-NN to cluster those datapoints, or use the "cells" of the space directly as clusters.

In any case, the SOM operator does not create a hierarchical clustering (and neither does k-NN), so you don't need the Flatten Clustering operator at all.

Best regards,
Marius

joe

Thank you for your help

joe

hi sir
may i ask you a question?
I am  afraid that the question is stupid ::)
How to import parameters to the epsilon and min points question in DBSCAN ? ???
May i use Set Macros model In order to impoet the parameters into DBSCAN?
very thanks
happy new year to you ;D

http://photo.xuite.net/chinchaowei/5389120/1.jpg

MariusHelf

Hi, from where do you want to import the parameters?

By the way, instead of the screenshots you can also post your processes directly by going to the XML tab in RapidMiner and copying the text there into your forum post (please use the "#" button on top of the input box to insert your xml code!)

Best,
Marius