Problem in saving incremental classifiers!

[palash247]

I am trying to train an incremental classifier and I am planning to update the classifier in future with new datasets as soon as they are available. For that, I need to save the trained incremental classifier. Saving a trained classifier can be done by using -d option but the option is not working in case of incremental classifiers. Following is the error I'm getting:

C:\Users\Palash\Dropbox\Project ME\java\meka-release-1.9.0>java -Xmx1024m -cp "./lib/*" meka.classifiers.multilabel.incremental.BRUpdateable -d classifier.dump -verbosity 5 -t train_dmoz_computers_dataset.arff -W weka.classifiers.meta.MOA -- -B moa.classifiers.trees.HoeffdingAdaptiveTree -output-debug-info

Evaluation exception (java.lang.Exception: Illegal options: -d classifier.dump ); failed to run experiment java.lang.Exception: Illegal options: -d classifier.dump at weka.core.Utils.checkForRemainingOptions(Utils.java:542) at meka.classifiers.multilabel.incremental.IncrementalEvaluation.evaluateModel(IncrementalEvaluation.java:83) at meka.classifiers.multilabel.incremental.IncrementalEvaluation.runExperiment(IncrementalEvaluation.java:45) at meka.classifiers.multilabel.incremental.BRUpdateable.main(BRUpdateable.java:72)

If saving classifier is not yet supported by incremental learners in meka, then how can I tweak it to get the job done.

[fracpete]

That wasn't supported, so I've added the -d/-l options to the incremental evaluation. You can download a snapshot with this functionality from here.
NB: Only minimal testing was performed.

[palash247]

Now getting the array index out of bound exception when I use option -d with following mentioned command. However, the same command executes completely if I don't use option -d. Following is the error:

[fracpete]

Copy/paste of text rather than screenshot would be appreciated. Retyping is prone to introducing typos.

I cannot reproduce the problem with the Music.arff dataset:

meka.classifiers.multilabel.incremental.PSUpdateable -t Music.arff -d classifier.dump -verbosity 5 -I 1000 -support 10 -P 0 -N 0 -W weka.classifiers.bayes.NaiveBayesUpdateable

However, I fixed a problem with the options: -S was already defined in the super class (for the seed value) and couldn't be used for the support. The option for supplying the support is now -support (as you can see in the above command-line string).