Package com.tibco.patterns.learn.training

Class COErrorRate.IterationResult

  • All Implemented Interfaces:
    java.lang.Cloneable
    Enclosing class:
    COErrorRate
    public static final class COErrorRate.IterationResult
extends java.lang.Object
implements java.lang.Cloneable
    Stores statistics for a full or partial training iteration

    • Constructor Detail

      • IterationResult

        public IterationResult​(int nFalseExamples,
                       int nTrueExamples,
                       int nErrors,
                       int nFalsePos,
                       int nFalseNeg,
                       int nUntrainedPred,
                       double totalConf)
        Stores all given values in the new object. Total confidence for errors and both total distance values are initialized to 0. Used only to store the values in this object (for Learn UI support). Do not use for model training.

      • IterationResult

        public IterationResult​(double threshold)
        Initializes all values to 0. Use only when calculating statistics for custom threshold and existing dataset, not for model training.
        Parameters:
        threshold - - a custom threshold between 0 and 1

    • Method Detail

      • clone

        public COErrorRate.IterationResult clone()
        Performs shallow copy of all simple types.
        Overrides:
        clone in class java.lang.Object

      • getNFalseExamples

        public int getNFalseExamples()
        Returns:
        number of accumulated examples with "False" label

      • getNTrueExamples

        public int getNTrueExamples()
        Returns:
        number of accumulated examples with "True" label

      • getNExamples

        public int getNExamples()
        Returns:
        total number of accumulated examples

      • getNErrors

        public int getNErrors()
        Returns:
        number of encountered errors. Includes errors with 0 prediction confidence.

      • getNFalsePos

        public int getNFalsePos()
        Returns:
        number of false positives, including the ones with 0 prediction confidence.

      • getNFalseNeg

        public int getNFalseNeg()
        Returns:
        number of false negatives, including the ones with 0 prediction confidence.

      • getNUntrainedPred

        public int getNUntrainedPred()
        Returns the number of examples that cannot be reliably predicted. Such examples have 0 prediction confidence. This happens when a submodel that predicts such example is completely untrained, or feature scores of the example are in an untrained or completely contradictory region of the submodel. More examples from the same subset and with similar feature scores should be added to the training dataset before using the trained model.
        Returns:
        number of examples that are predicted with 0 confidence.

      • getTotalConf

        public double getTotalConf()
        Returns:
        total confidence for all examples (sum of all confidence values).

      • errorRate

        public double errorRate()
        Returns:
        error rate. Includes errors with 0 prediction confidence.

      • falsePosRate

        public double falsePosRate()
        Returns:
        false positive rate. Includes false positives with 0 prediction confidence.

      • falseNegRate

        public double falseNegRate()
        Returns:
        false negative rate. Includes false negatives with 0 prediction confidence.

      • noPredictionRate

        public double noPredictionRate()
        Returns the percentage of examples that cannot be reliably predicted. See getNUntrainedPred() for details about such examples.
        Returns:
        untrained prediction rate - the percentage of examples that have 0 prediction confidence.

      • meanConf

        public double meanConf()
        Returns:
        mean prediction confidence for all examples

      • meanErrConf

        public double meanErrConf()
        Returns:
        mean prediction confidence for incorrectly classified examples

      • meanDist

        public double meanDist()
        Returns:
        mean distance between the model prediction score and the actual label (0 or 1) for all examples

      • meanErrDist

        public double meanErrDist()
        Returns:
        mean distance between the model prediction score and the actual label (0 or 1) for incorrectly classified examples

      • isConverged

        public boolean isConverged​(double targetErrorRate)
        Returns true if the target error rate has been reached. The error rate includes the errors for examples predicted with 0 confidence (see getNUntrainedPred() for more details). Some of these errors may be from submodels that are never trained, so model training will have no impact for these errors.
        Returns:
        true if the error rate of the last training iteration is below the target error rate.

      • isConverged

        public boolean isConverged()
        Returns true if the default target error rate has been reached. See isConverged(double).
        Returns:
        true if the error rate of the last training iteration is below the default target error rate.

      • isPerfectResult

        public boolean isPerfectResult()
        Returns:
        true if all examples in the dataset were predicted correctly.

      • addPrediction

        public void addPrediction​(RLinkOut prediction,
                          boolean label)
        Updates all stored statistics based on the additional model prediction.
        Parameters:
        prediction - - the new model prediction
        label - - the actual label that had to be predicted