csli.util.classify.stanford
Class ClassifierUtils

java.lang.Object
  csli.util.classify.stanford.ClassifierUtils

public abstract class ClassifierUtils
extends Object

Constructor Summary

ClassifierUtils()

Method Summary

static <D extends edu.stanford.nlp.dbm.Datum> Map<Integer,Double>	getFeatures(D datum, Map<String,Integer> featureMap)
static Pair<String,Double>	getValuedFeature(Object feature)
static <D extends edu.stanford.nlp.dbm.Datum> List<Pair<Double,Double>>	normalizeFeatures(List<D> examples, Map<String,Integer> featureMap)
static double	pLogP(double p) Convenience method to get round the fact that multiplying zero by -Inf gives NaN
static <D extends edu.stanford.nlp.dbm.Datum> void	prune(List<D> examples, Set<String> toPrune)
static <D extends edu.stanford.nlp.dbm.Datum> Set<String>	pruneFeatures(List<D> examples, List<String> specs, String posLabel, String negLabel)
static <D extends edu.stanford.nlp.dbm.Datum> Set<String>	pruneFeaturesByCorrelation(List<D> examples, double margin, String posLabel, String negLabel)
static <D extends edu.stanford.nlp.dbm.Datum> Set<String>	pruneFeaturesByFreq(List<D> examples, double threshold)
static <D extends edu.stanford.nlp.dbm.Datum> Set<String>	pruneFeaturesByInfoGain(List<D> examples, double margin, String posLabel, String negLabel)

Methods inherited from class java.lang.Object

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

Constructor Detail

ClassifierUtils

public ClassifierUtils()

Method Detail

getFeatures

public static <D extends edu.stanford.nlp.dbm.Datum> Map<Integer,Double> getFeatures(D datum,
                                                                                     Map<String,Integer> featureMap)

Type Parameters:

D - a subclass of Datum

Parameters:

datum - the data instance

featureMap - a map of feature label to feature index number, or null to assume ordered numeric features

Returns:

a map from feature index number to feature value

getValuedFeature

public static Pair<String,Double> getValuedFeature(Object feature)

Parameters:

feature - a feature, either as a ScoredObject or just a plain Object

Returns:

a Pair of the String feature and its Double value (the score of a ScoredObject, or 1.0 otherwise)

normalizeFeatures

public static <D extends edu.stanford.nlp.dbm.Datum> List<Pair<Double,Double>> normalizeFeatures(List<D> examples,
                                                                                                 Map<String,Integer> featureMap)

Type Parameters:

D - a subclass of Datum

Parameters:

examples - a List of data instances to be normalized

featureMap - a map of feature label to feature number, or null to assume ordered numeric features

Returns:

a List of Pairs of Doubles which record the normalization: f_norm = (f_raw-a)*b

pruneFeatures

public static <D extends edu.stanford.nlp.dbm.Datum> Set<String> pruneFeatures(List<D> examples,
                                                                               List<String> specs,
                                                                               String posLabel,
                                                                               String negLabel)

Type Parameters:

D -

Parameters:

examples - a list of Datum instances from which to prune features

specs - a list of String pruning specifications

posLabel -

negLabel -

Returns:

the set of feature labels removed

pruneFeaturesByFreq

public static <D extends edu.stanford.nlp.dbm.Datum> Set<String> pruneFeaturesByFreq(List<D> examples,
                                                                                     double threshold)

Type Parameters:

D -

Parameters:

examples - a list of Datum instances from which to prune features

threshold - the value of frequency below which features will be pruned

Returns:

the set of feature labels removed

pruneFeaturesByCorrelation

public static <D extends edu.stanford.nlp.dbm.Datum> Set<String> pruneFeaturesByCorrelation(List<D> examples,
                                                                                            double margin,
                                                                                            String posLabel,
                                                                                            String negLabel)

Type Parameters:

D -

Parameters:

examples - a list of Datum instances from which to prune features

posLabel -

negLabel -

Returns:

the set of feature labels removed

pruneFeaturesByInfoGain

public static <D extends edu.stanford.nlp.dbm.Datum> Set<String> pruneFeaturesByInfoGain(List<D> examples,
                                                                                         double margin,
                                                                                         String posLabel,
                                                                                         String negLabel)

Type Parameters:

D -

Parameters:

examples - a list of Datum instances from which to prune features

posLabel -

negLabel -

Returns:

the set of feature labels removed

prune

public static <D extends edu.stanford.nlp.dbm.Datum> void prune(List<D> examples,
                                                                Set<String> toPrune)

Type Parameters:

D -

Parameters:

examples - a list of Datum instances from which to prune features

toPrune - a set of feature labels to remove

pLogP

public static double pLogP(double p)

Convenience method to get round the fact that multiplying zero by -Inf gives NaN

Parameters:

p - a double probability

Returns:

p*Math.log(p), which will be 0 if p=0