FastNominalAttributeClassObserver implemented

* improves performance for nominal attributes with large number
of attribute values, it simply calculates distribution on the
fly to avoid loops in getClassDistsResultingFrom* and probabilityOfAttributeValueGivenClass methods.

moa/src/main/java/moa/classifiers/core/attributeclassobservers/FastNominalAttributeClassObserver.java

@@ -0,0 +1,173 @@
+/*
+ *    FastNominalAttributeClassObserver.java
+ *    Copyright (C) 2007 University of Waikato, Hamilton, New Zealand
+ *    @author Richard Kirkby ([email protected])
+ *
+ *    This program is free software; you can redistribute it and/or modify
+ *    it under the terms of the GNU General Public License as published by
+ *    the Free Software Foundation; either version 3 of the License, or
+ *    (at your option) any later version.
+ *
+ *    This program is distributed in the hope that it will be useful,
+ *    but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *    GNU General Public License for more details.
+ *
+ *    You should have received a copy of the GNU General Public License
+ *    along with this program. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+package moa.classifiers.core.attributeclassobservers;
+
+import moa.classifiers.core.AttributeSplitSuggestion;
+import moa.classifiers.core.conditionaltests.NominalAttributeBinaryTest;
+import moa.classifiers.core.conditionaltests.NominalAttributeMultiwayTest;
+import moa.classifiers.core.splitcriteria.SplitCriterion;
+import moa.core.DoubleVector;
+import moa.core.ObjectRepository;
+import moa.core.Utils;
+import moa.options.AbstractOptionHandler;
+import moa.tasks.TaskMonitor;
+
+import java.util.HashMap;
+import java.util.Map;
+
+
+/**
+ * Class for observing the class data distribution for a nominal attribute.
+ * This observer monitors the class distribution of a given attribute.
+ * Used in naive Bayes and decision trees to monitor data statistics on leaves.
+ *
+ * @author Richard Kirkby ([email protected])
+ * @author Eugene Kamenev ([email protected])
+ * @version $Revision: 7 $
+ */
+public class FastNominalAttributeClassObserver extends AbstractOptionHandler implements DiscreteAttributeClassObserver {
+
+    private static final long serialVersionUID = 1L;
+
+    protected double totalWeightObserved = 0.0;
+
+    protected double missingWeightObserved = 0.0;
+
+    protected Map<Integer, Map<Integer, Double>> attValDistPerClassCount = new HashMap<>();
+
+    protected Map<Integer, Double> classTotalCount = new HashMap<>();
+
+    protected Map<Integer, Integer> maxAttrValue = new HashMap<>();
+
+    @Override
+    public void observeAttributeClass(double attVal, int classVal, double weight) {
+        if (Utils.isMissingValue(attVal)) {
+            this.missingWeightObserved += weight;
+        } else {
+            int attValInt = (int) attVal;
+            Map<Integer, Double> valDistCount = this.attValDistPerClassCount.computeIfAbsent(classVal, k -> new HashMap<>());
+            // update distribution count
+            valDistCount.put(attValInt, valDistCount.getOrDefault(attValInt, 0.0) + weight);
+            Integer maxValue = this.maxAttrValue.get(classVal);
+            if (maxValue == null || attVal > maxValue) {
+                // update max attribute value
+                this.maxAttrValue.put(classVal, attValInt);
+            }
+            // update the total count for the class
+            this.classTotalCount.put(classVal, this.classTotalCount.getOrDefault(classVal, 0.0) + weight);
+        }
+        this.totalWeightObserved += weight;
+    }
+
+    @Override
+    public double probabilityOfAttributeValueGivenClass(double attVal, int classVal) {
+        Map<Integer, Double> obs = this.attValDistPerClassCount.get(classVal);
+        Double sumCounts = this.classTotalCount.getOrDefault(classVal, 0.0);
+        Integer max = this.maxAttrValue.getOrDefault(classVal, (int) attVal) + 1;
+        return obs != null ? (obs.getOrDefault((int) attVal, 0.0) + 1.0) / (sumCounts + max) : 0.0;
+    }
+
+    public double[][] getClassDistsResultingFromMultiwaySplit(int maxAttValsObserved) {
+        DoubleVector[] resultingDists = new DoubleVector[maxAttValsObserved];
+        for (int i = 0; i < resultingDists.length; i++) {
+            resultingDists[i] = new DoubleVector();
+        }
+        for (Map.Entry<Integer, Map<Integer, Double>> entry : this.attValDistPerClassCount.entrySet()) {
+            int classVal = entry.getKey();
+            Map<Integer, Double> attValDistCount = entry.getValue();
+            for (int j = 0; j < maxAttValsObserved; j++) {
+                resultingDists[j].addToValue(classVal, attValDistCount.getOrDefault(j, 0.0));
+            }
+        }
+        double[][] distributions = new double[maxAttValsObserved][];
+        for (int i = 0; i < distributions.length; i++) {
+            distributions[i] = resultingDists[i].getArrayRef();
+        }
+        return distributions;
+    }
+
+    public double[][] getClassDistsResultingFromBinarySplit(int valIndex) {
+        DoubleVector equalsDist = new DoubleVector();
+        DoubleVector notEqualDist = new DoubleVector();
+        for (Map.Entry<Integer, Map<Integer, Double>> entry : this.attValDistPerClassCount.entrySet()) {
+            int classVal = entry.getKey();
+            Map<Integer, Double> attValDistCount = entry.getValue();
+            double count = attValDistCount.getOrDefault(valIndex, 0.0);
+            equalsDist.addToValue(classVal, count);
+            notEqualDist.addToValue(classVal, this.classTotalCount.get(classVal) - count);
+        }
+        return new double[][]{equalsDist.getArrayRef(),
+                notEqualDist.getArrayRef()};
+    }
+
+    @Override
+    public AttributeSplitSuggestion getBestEvaluatedSplitSuggestion(SplitCriterion criterion, double[] preSplitDist,
+                                                                    int attIndex, boolean binaryOnly) {
+        AttributeSplitSuggestion bestSuggestion = null;
+        int maxAttValsObserved = 0;
+        for (Integer max : this.maxAttrValue.values()) {
+            if (max > maxAttValsObserved) {
+                maxAttValsObserved = max + 1;
+            }
+        }
+        if (!binaryOnly) {
+            double[][] postSplitDists = getClassDistsResultingFromMultiwaySplit(maxAttValsObserved);
+            double merit = criterion.getMeritOfSplit(preSplitDist,
+                    postSplitDists);
+            bestSuggestion = new AttributeSplitSuggestion(
+                    new NominalAttributeMultiwayTest(attIndex), postSplitDists,
+                    merit);
+        }
+        for (int valIndex = 0; valIndex < maxAttValsObserved; valIndex++) {
+            double[][] postSplitDists = getClassDistsResultingFromBinarySplit(valIndex);
+            double merit = criterion.getMeritOfSplit(preSplitDist,
+                    postSplitDists);
+            if ((bestSuggestion == null) || (merit > bestSuggestion.merit)) {
+                bestSuggestion = new AttributeSplitSuggestion(
+                        new NominalAttributeBinaryTest(attIndex, valIndex),
+                        postSplitDists, merit);
+            }
+        }
+        return bestSuggestion;
+    }
+
+    @Override
+    public void observeAttributeTarget(double v, double v1) {
+        throw new UnsupportedOperationException("Not supported yet.");
+    }
+
+    @Override
+    protected void prepareForUseImpl(TaskMonitor monitor, ObjectRepository repository) {
+        // TODO Auto-generated method stub
+    }
+
+    @Override
+    public void getDescription(StringBuilder sb, int indent) {
+        // TODO Auto-generated method stub
+    }
+
+    public double totalWeightOfClassObservations() {
+        return this.totalWeightObserved;
+    }
+
+    public double weightOfObservedMissingValues() {
+        return this.missingWeightObserved;
+    }
+}