Package org.pentaho.di.scoring

Class WekaScoringModel

  • All Implemented Interfaces:
    Serializable
    public abstract class WekaScoringModel
extends Object
implements Serializable
    Abstract wrapper class for a Weka model. Provides a unified interface to obtaining predictions. Subclasses ( WekaScoringClassifer and WekaScoringClusterer) encapsulate the actual weka models.
    Version:
    1.0
    Author:
    Mark Hall (mhall{[at]}pentaho.org)
    See Also:
    Serialized Form

    • Constructor Detail

      • WekaScoringModel

        public WekaScoringModel​(Object model)
        Creates a new WekaScoringModel instance.
        Parameters:
        model - the actual Weka model to enacpsulate

    • Method Detail

      • setLog

        public void setLog​(org.pentaho.di.core.logging.LogChannelInterface log)
        Set the log to pass on to the model. Only PMML models require logging.
        Parameters:
        log - the log to use

      • setHeader

        public void setHeader​(weka.core.Instances header)
        Set the Instances header
        Parameters:
        header - an Instances value

      • getHeader

        public weka.core.Instances getHeader()
        Get the header of the Instances that was used build this Weka model
        Returns:
        an Instances value

      • done

        public void done()
        Tell the model that this scoring run is finished.

      • setModel

        public abstract void setModel​(Object model)
        Set the weka model
        Parameters:
        model - the Weka model

      • getModel

        public abstract Object getModel()
        Get the weka model
        Returns:
        the Weka model as an object

      • classifyInstance

        public abstract double classifyInstance​(weka.core.Instance inst)
                                 throws Exception
        Return a classification. What this represents depends on the implementing sub-class. It could be the index of a class-value, a numeric value or a cluster number for example.
        Parameters:
        inst - the Instance to be classified (predicted)
        Returns:
        the prediction
        Throws:
        Exception - if an error occurs

      • distributionForInstance

        public abstract double[] distributionForInstance​(weka.core.Instance inst)
                                          throws Exception
        Return a probability distribution (over classes or clusters).
        Parameters:
        inst - the Instance to be predicted
        Returns:
        a probability distribution
        Throws:
        Exception - if an error occurs

      • classifyInstances

        public abstract double[] classifyInstances​(weka.core.Instances insts)
                                    throws Exception
        Batch scoring method. Call isBatchPredictor() first in order to determine if the underlying model can handle batch scoring.
        Parameters:
        insts - the instances to score
        Returns:
        an array of predictions
        Throws:
        Exception - if a problem occurs

      • distributionsForInstances

        public abstract double[][] distributionsForInstances​(weka.core.Instances insts)
                                              throws Exception
        Batch scoring method. Call isBatchPredictor() first in order to determine if the underlying model can handle batch scoring.
        Parameters:
        insts - the instances to score
        Returns:
        an array of probability distributions, one for each instance
        Throws:
        Exception - if a problem occurs

      • isSupervisedLearningModel

        public abstract boolean isSupervisedLearningModel()
        Returns true if the encapsulated Weka model is a supervised model (i.e. has been built to predict a single target in the data).
        Returns:
        true if the encapsulated Weka model is a supervised model

      • isUpdateableModel

        public abstract boolean isUpdateableModel()
        Returns true if the encapsulated Weka model can be updated incrementally in an instance by instance fashion.
        Returns:
        true if the encapsulated Weka model is incremental model

      • isBatchPredictor

        public abstract boolean isBatchPredictor()
        Returns true if the encapsulated Weka model can produce predictions in a batch.
        Returns:
        true if the encapsulated Weka model can produce predictions in a batch

      • update

        public abstract boolean update​(weka.core.Instance inst)
                        throws Exception
        Update (if possible) a model with the supplied Instance
        Parameters:
        inst - the Instance to update the model with
        Returns:
        true if the model was updated
        Throws:
        Exception - if an error occurs

      • createScorer

        public static WekaScoringModel createScorer​(Object model)
                                     throws Exception
        Static factory method to create an instance of an appropriate subclass of WekaScoringModel given a Weka model.
        Parameters:
        model - a Weka model
        Returns:
        an appropriate WekaScoringModel for this type of Weka model
        Throws:
        Exception - if an error occurs