Package com.tibco.patterns.learn.api.hint

Class HintAddUnderrepr

java.lang.Object

com.tibco.patterns.learn.api.hint.HintAddPairs

com.tibco.patterns.learn.api.hint.HintAddUnderrepr

All Implemented Interfaces:

Hint

public final class HintAddUnderrepr
extends HintAddPairs

Hint to add pairs to subsets that are underrepresented in training dataset.

Field Summary

Fields inherited from class com.tibco.patterns.learn.api.hint.HintAddPairs

subsets

Method Summary

Modifier and Type	Method	Description
static HintAddUnderrepr	create(FeatureQuery fq, SubsetFamily trainSubsets, double nStdDevBelowMean, int minNPairs)	Creates a hint object with all underrepresented subsets, selecting them from subsets in training dataset.

Methods inherited from class com.tibco.patterns.learn.api.hint.HintAddPairs

getCount, getSubsets, size, suitableRecord, toString

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Method Detail

create

public static HintAddUnderrepr create(FeatureQuery fq,
                                      SubsetFamily trainSubsets,
                                      double nStdDevBelowMean,
                                      int minNPairs)

Creates a hint object with all underrepresented subsets, selecting them from subsets in training dataset. Subset is underrepresented if the number of pairs for that subset in training dataset is:
1. < minNPairs OR
2. < (mean number of pairs) - nStdDevBelowMean * (standard deviation of number of pairs), if standard deviation is small enough
3. < a number that approaches the mean if standard deviation is very large
Assumes a trained model was saved (uses feature scores).

Parameters:

trainSubsets - - all subsets in training dataset. No need to search validation dataset - it is already analyzed for untrained subsets

nStdDevBelowMean - - the limit (expressed in standard deviations below mean)

minNPairs - - minimum number of pairs required in training dataset for any subset.

Returns:

the new hint object. Returns null if no underrepresented subsets can be found (or model was never trained).

Throws:

java.lang.IllegalArgumentException - if any of the parameters are negative.