What's new in Weka 3.8.1 and 3.9.1

In core weka:

  • Package manager now handles redirects generated by SourceForge
  • Package manager now employs a new class loading mechanism that attempts to avoid third-party library clashes by isolating the third-party libraries in each package
  • new RelationNameModifier, SendToPerspective, WriteWekaLog, Job, StorePropertiesInEnvironment, SetPropertiesFromEnvironment, WriteDataToResult and GetDataFromResult steps in Knowledge Flow
  • RandomForest now has an option for computing the mean impurity decrease variable importance scores
  • JRip now prunes redundant numeric attribute-value tests from rules
  • Knowledge Flow now offers an additional executor service that uses a single worker thread; steps can, if necessary, declare programatically that they should run in the single-threaded executor.
  • GUIs with result lists now support multi-entry delete
  • GUIs now support copying/pasting of array configurations to/from the clipboard

In packages:

  • Multi-class FLDA in the discriminantAnalysis package
  • New implementations in the ensemblesOfNestedDichotomies package
  • distributedWekaBase now includes the latest version of Ted Dunning's t-digest quantile estimator, bringing a factor of 4 speedup over the old implementation
  • New streamingUnivariateStats package
  • RPlugin package updated to support the latest version of MLR
  • New wekaDeepLearning4j package - provides a MLP classifier built using the DL4J library. Can work with either CPU-based or GPU-based native libraries
  • New logarithmicErrorMetrics package
  • New RankCorrelation package, courtesy of Quan Sun. Provides rank correlation metrics, Kendall tau and Spearman rho, for evaluating regression schemes
  • New AffectiveTweets package, courtesy of Felipe Bravom. Provides text filters for sentiment analysis of tweets
  • New AnalogicalModeling package, courtesy of Nathan Glenn. Provides an exemplar-based approach to modeling
  • New MultiObjectiveEvolutionaryFuzzyClassifier package, courtesy of Carlos Martinez Cortes. Provides a fuzzy rule-based classifier
  • New MultiObjectiveEvolutionarySearch package, courtesy of Carlos Martinez Cortes. Provides a search method that uses the ENORA multi-objective evolutionary algorithm