McDiarmid Drift Detection Methods

This repository is prepared for McDiarmid Drift Detection Methods.

• The source codes of MDDMs and FHDDM are available in repository codes_for_moa.
  • The source codes should be added into the MOA framework, under folder 'moa/classifiers/core/driftdetection/'.
  • You may need to install Eclipse (or a similar Java IDE), and then import MOA as a Maven project.
  • For further informaiton, please click here for the manual of MOA.
• The synthetic data streams are available in the data_streams repository, (you need to unzip the files).
• You also need to pass the path of sizeofag.jar as a virual machine (VM) argument for being able to measure memory size of objects.

  -javaagent:[enter directory path]/sizeofag.jar
  

  For Eclipse, you must pass the path using tab Arguments in Run Configurations... under tab Run from the main menu.
• I have already built a moa.jar file and shared it in this repository. You should run it in a terminal (or in a command prompt in Windows). You may use a command similar to the one below for an experiment:

  java -cp moa.jar -javaagent:sizeofag.jar moa.DoTask alipsgh.EvaluatePrequential -l \(drift.alipsgh.DriftDetectionMethodClassifierSyntheticData -l trees.HoeffdingTree -d MDDM_G -m 20000 -n 4 -a 250\) -s \(ArffFileStream -f /home/ali/moa/data_streams/arff/sine1_w_50_n_0.1/sine1_w_50_n_0.1_101.arff\) -e BasicClassificationPerformanceEvaluator -i -1 -f 1000 -q 1000
  

  The results would be similar to:

  Delay, TP, FP, FN, Runtime, Memory, Classification Accuracy
  39.25,4,0,0,49,1344,87.05
  

Please contact me for any problem.

_{Ali Pesaranghader © 2018}