Using Weka


Now, it is easy to use MOA classifiers and streams from WEKA, and WEKA classifiers from MOA.
The main difference between using incremental classifiers in WEKA and in MOA will be the evaluation method used.
Weka classifiers may be incremental or non incremental methods.
To use the Weka classifiers from MOA it is necessary to use one of the following classes:

  • WEKAClassifier: A classifier to use classifiers from WEKA.
  •   WEKAClassifier -l weka.classifiers.trees.J48
                -w 10000 -i 1000 -f 100000

    Parameters:
    -l : Classifier to train
    -w : Size of Window for training learner
    -i : Size of first Window for training learner
    -f : How many instances between model builds

  • SingleClassifierDrift: Class for handling concept drift datasets with a wrapper on a classifier.
  •  SingleClassifierDrift -d EDDM
        -l weka.classifiers.bayes.NaiveBayesUpdateable

    Parameters:
    -l : Classifier to train
    -d : Drift detection method to use: DDM or EDDM

You can use MOA classifiers quite easily as incremental classifiers within the WEKA Explorer, Knowledge Flow interface or command-line interface, using the weka.classifiers.meta.MOA meta-classifier. This meta-classifier is just a wrapper for MOA classifiers, translating the WEKA method calls into MOA ones.