"DBSCAN Error: IndexoutofBoundexception"

shadrigo

hi,

I am not very familiar with rapidminer(yet!!) so I hope my problem is just a failure of a beginner who hasn´t found where to find the
right solution or gets a feeling for the error source.

I want to use DBSCAN Clustering in order to cluster image histograms with values for each Bin between 0.0 and 100 %.
My example source is an arff File.
The process is set up with the examplesource and then follows a NominalAttributeFilter before the exampleset is put into
DBSCAN.
I get an IndexOutOfBoundsException failure with the Message "IndexOutOfBoundsException caught: Index: 0 , Size: 0".

In Weka I don´t get such an error, so I think my example data seems to be valid.
Any suggestions? Thank you in advance!

land

Hi,
unfortunatly you are right. There was a small bug, causing this exception if the parameters have been set to values which result in classifying all examples as noise.
This is now fixed in the repository (head). Nevertheless no sensefull results will be delivered, since every example is classified as noise.

DBScan is one of the most instable clustering algorithms I have seen so far, so tuning of parameter epsilon is very important to get a usefull result.

Greetings,
  Sebastian

shadrigo

thanks for your answer.

I will check out from the repository and try again but will consider your further hints on dbscan so that I play a little bit with the parameters but
perhaps look for an alternative for clustering.

Greets
Thomas