CODEWITHSUNDEEP
weka
pedrosaurio
pedrosaurio

Reputation: 4926

how to ignore list of attributes using the command line while clustering in weka?

I am running a series of clustering analyses in weka and I have realized that automatizing it is the way to go if I want to get somewhere. I'll explain a bit how I am working.

I am trying to reproduce some results I've got by using the GUI, but now with commands in the CLI. The problem is that I usually ignore a list of attributes when clustering using the GUI. I cannot find a way of selecting a list of attributes to be ignored in the command line.

For example:

java weka.clusterers.XMeans \
-I 10 -M 1000 -J 1000 \
-L 2 -H 9 -B 1.0 -C 0.25 \
-D "weka.core.MinkowskiDistance -R first-last" -S 10 \
-t "/home/pedrosaurio/bigtable.arff"

My experience with weka is limited so I don't know if I am missing some basic understanding of how it works.

Upvotes: 4

Views: 2874

Answers (2)

A Clemotte
A Clemotte

Reputation: 61

To ignore an attribute you have to do it from the distance function

Ignore attributes from command line (Matlab):

COLUMNS = '3-last'; % The indices start from 1, 'first' and 'last' are valid as well. E.g .: first-3,5,6-last
Df = weka.core.EuclideanDistance (); % Setup distance function.
Df.setAttributeIndices (COLUMNS); % Setup distance function.

Ignore attributes from GUI Ignore attributes from GUI

I do not understand why when someone asks how to ignore attributes all the answers say how to modify the dataset, using a filter in the preprocess section.

Upvotes: 0

Atilla Ozgur
Atilla Ozgur

Reputation: 14721

Data Preprocessing functions are called filters. You need to use filters together with cluster algorithm. See below example.

java weka.clusterers.FilteredClusterer \ 
-F weka.filters.unsupervised.attribute.Remove -V -R 1,5  \
-W weka.clusterers.XMeans  -I 10 -M 1000   -J 1000  -L 2 -H 9 -B 1.0 -C 0.25 \ 
-D "weka.core.MinkowskiDistance -R first-last" -S 10 \ 
-t "/home/pedrosaurio/bigtable.arff"

Here we remove attributes 1-5 then use xmeans.

Upvotes: 5

Related Questions