from MOA analysis to a real app

I was able to compare many weka classifiers on my target streams and now I am wondering how to use sliding window and drift detection in my real app. I plan to use non incremental classifiers so I need to keep the right dataset from my stream considering the expected sliding window and drift detection management. How MOA or Weka provide the required information (old instances to discard, optimal number of instances to keep) ?

Thank you,
Xavier