Hoffman2:WEKA
WEKA is a machine learning program built by the Kiwis down in New Zealand. It has an extensive library of machine learning algorithms.
WEKA
Getting Started with WEKA
Introductory material and full manuals can be found here.
There are sample data sets that can be found on Hoffman2 at
/u/project/CCN/apps/weka/current/data
and most of these are used in the introductory materials.
GUI
To run the WEKA GUI from Hoffman2
- SSH into Hoffman2 making sure you have X11 forwarding enabled ("-X" flag)
- Check out an interactive computing node
- Run the command
$ java -jar /u/project/CCN/apps/weka/current/weka.jar
- And the current version of WEKA should start up and the GUI will appear.
Command Line
Calling any of the usual WEKA command line tools will work in submitted jobs so long as you properly reference the WEKA jar file at:
/u/project/CCN/apps/weka/current/weka.jar
A sample command would be
$ java -Xmx2g -cp /u/project/CCN/apps/weka/current/weka.jar weka.classifiers.functions.SMO -o -t awesome_train_file.arff -T awesome_test_file.arff
where
- -Xmx2g
- calls for 2GB of memory
- -Xmx256m calls for 256MB of memory
- -o
- suppresses the model output
- -t awesome_train_file.arff
- specifies the training file
- -T awesome_test_file.arff
- is an optional testing file for the model
- if this option is not provided, 10 fold cross validation will be done using Training data
Massive Online Analysis (MOA)
MOA is an extension of WEKA to help with larger datasets.
GUI
To run the MOA GUI from Hoffman2
- SSH into Hoffman2 making sure you have X11 forwarding enabled ("-X" flag)
- Check out an interactive computing node
- Run the command
$ moa.sh
- The current version of MOA should start up and the GUI will appear.
Command Line
Calling any of the usual MOA command line tools will work in submitted jobs so long as you properly reference the MOA jar file at:
/u/project/CCN/apps/weka/current/weka.jar
A sample command would be
$ java -Xmx4g -cp /u/project/CCN/apps/moa/current/moa.jar -javaagent:sizeofag.jar moa.DoTask "EvaluatePrequential -l HoeffdingTree -i 1000000 -w 10000"
where
- -Xmx4g
- calls for 4GB of memory
- -Xmx256m calls for 256MB of memory
- EvaluatePrequential -l HoeffdingTree -i 1000000 -w 10000
- is the MOA command to execute