Name

adams.flow.transformer.WekaClassifierRanker

Synopsis

Performs a quick evaluation using cross-validation on a single dataset (or evaluation on a separate test set if the number of folds is less than 2) to rank the classifiers received on the input and forwarding the x best ones. Further evaluation can be performed using the Experimenter.

Additional information

Flow input/output:
- input: weka.classifiers.Classifier[]
- output: weka.classifiers.Classifier[]

Options

• loggingLevel
  

  The logging level for outputting errors and debugging output.

  command-line	-logging-level <OFF|SEVERE|WARNING|INFO|CONFIG|FINE|FINER|FINEST>
  default	WARNING
  min-user-mode	Expert

  
  
• name
  

  The name of the actor.

  command-line	-name <java.lang.String>
  default	WekaClassifierRanker

  
  
• annotations
  

  The annotations to attach to this actor.

  command-line	-annotation <adams.core.base.BaseAnnotation>
  default

  
  
• skip
  

  If set to true, transformation is skipped and the input token is just forwarded as it is.

  command-line	-skip <boolean>
  default	false

  
  
• stopFlowOnError
  

  If set to true, the flow execution at this level gets stopped in case this actor encounters an error; the error gets propagated; useful for critical actors.

  command-line	-stop-flow-on-error <boolean>
  default	false
  min-user-mode	Expert

  
  
• silent
  

  If enabled, then no errors are output in the console; Note: the enclosing actor handler must have this enabled as well.

  command-line	-silent <boolean>
  default	false
  min-user-mode	Expert

  
  
• max
  

  The maximum number of top-ranked classifiers to forward; use -1 to forward all of them (ranked array).

  command-line	-max <int>
  default	3
  minimum	-1

  
  
• seed
  

  The seed value to use in the cross-validation.

  command-line	-seed <long>
  default	1

  
  
• folds
  

  The number of folds to use in cross-validation.

  command-line	-folds <int>
  default	10
  minimum	1

  
  
• measure
  

  The measure used for ranking the classifiers.

  command-line	-measure <Correlation coefficient|Root mean squared error|Root relative squared error|Mean absolute error|Root absolute error|Combined: (1-abs(CC)) + RRSE + RAE|Accuracy|Kapp>
  default	CC

  
  
• classLabel
  

  The class label index to use in case of class-specific measures.

  command-line	-class-label <adams.data.weka.WekaLabelIndex>
  default	first
  example	An index is a number starting with 1; apart from label names (case-sensitive), the following placeholders can be used as well: first, second, third, last_2, last_1, last; numeric indices can be enforced by preceding them with '#' (eg '#12'); label names can be surrounded by double quotes.

  
  
• train
  

  The name of the callable actor that is used for obtaining the training set.

  command-line	-train <adams.flow.core.CallableActorReference>
  default	train

  
  
• test
  

  The name of the callable actor that is used for obtaining the test set (only if folds <2).

  command-line	-test <adams.flow.core.CallableActorReference>
  default	test

  
  
• outputBestSetup
  

  If true, then for optimizers like GridSearch and MultiSearch the best setup that was found will be output instead of the optimizer setup.

  command-line	-output-best <boolean>
  default	false

  
  
• numThreads
  

  The number of threads to use for evaluating the classifiers in parallel (-1 means one for each core/cpu).

  command-line	-num-threads <int>
  default	0