DKPro TC

Discriminators (0.6.0 Release)

Mandatory Parameters

  • readers (DimensionBundle)
    • readerTrain: name of the reader class for the training data
    • readerTrainParams: a list of parameterName-value pairs to configure the training data reader
    • readerTest: name of the reader class for the test data (not necessary for Crossvalidation)
    • readerTestParams: a list of parameterName-value pairs to configure the test data reader (not necessary for Crossvalidation)
  • featureSet: a list of feature extractor class names (the feature extractors to use)
  • pipelineParameters: a list of parameterName-value pairs (parameters necessary to configure the feature extractors)
  • dataWriter: a DataWriter class (a writer to produce the input for the classification framework, e.g. Weka)
  • classificationArguments: a Weka/Meka classifier class and list of arguments to parametrize it (the classification algorithm)
  • featureMode: one of document, unit, pair, or sequence
  • learningMode: one of singleLabel, multiLabel, or regression

Optional Parameters

  • threshold: boolean (the threshold to create a bipratition from a ranking; only for multiLabel learning mode)
  • featureSelection (DimensionBundle)
    • attributeEvaluator: Weka attribute selection evaluation class and list of arguments to parametrize it (the attribute evaluation algorithm)
    • featureSearcher: Weka attribute selection search class and list of arguments to parametrize it (the ranking algorithm; only for singleLabel learning mode)
    • labelTransformationMethod: a Mulan label transformation method (the label transformation method; only for multiLabel learning mode)
    • numLabelsToKeep: integer (the number of features which will be selected; only for multiLabel learning mode)
    • applySelection: boolean (whether to actually apply the )
  • developerMode: boolean (if true, you will not be warned when using feature extractors incompatible with the specified featureMode)