Nominal attribute values not present in the data not considered in the induction.

Author: agudys

adaa.analytics.rules/build.gradle

@@ -5,7 +5,7 @@ plugins {
     id 'java'
 }
 
-version = '2.1.18b'
+version = '2.1.19'
 java {
     sourceCompatibility = JavaVersion.VERSION_1_8
 }

adaa.analytics.rules/src/main/java/adaa/analytics/rules/logic/induction/AbstractFinder.java

@@ -58,8 +58,8 @@ public abstract class AbstractFinder implements AutoCloseable {
 
 	private List<IFinderObserver> observers = new ArrayList<IFinderObserver>();
 
-	protected Map<IAttribute, Integer[]> attributeValuesOrder
-			= new HashMap<IAttribute, Integer[]>();
+	protected Map<IAttribute, List<Integer>> attributeValuesOrder
+			= new HashMap<IAttribute, List<Integer>>();
 
 	public void addObserver(IFinderObserver o) { observers.add(o); }
 	public void clearObservers() { observers.clear(); }
@@ -94,17 +94,21 @@ public void preprocess(IExampleSet trainSet) {
 
 		for (IAttribute attr : attributes) {
 
-			Integer[] valuesOrder = null;
+			List<Integer> valuesOrder = null;
 
 			// check if attribute is nominal
 			if (attr.isNominal()) {
 				// get orders
-				valuesOrder = new Integer[attr.getMapping().size()];
+				valuesOrder = new ArrayList<Integer>();
 				List<String> labels = new ArrayList<>();
 				labels.addAll(attr.getMapping().getValues());
 				Collections.sort(labels);
 				for (int j = 0; j < labels.size(); ++j) {
-					valuesOrder[j] = attr.getMapping().getIndex(labels.get(j));
+					int index = attr.getMapping().getIndex(labels.get(j));
+
+					if (trainSet.getDoubleColumn(attr).contains((double)index)) {
+						valuesOrder.add(index);
+					}
 				}
 			}
 

adaa.analytics.rules/src/main/java/adaa/analytics/rules/logic/induction/ApproximateClassificationFinder.java

@@ -104,12 +104,16 @@ public void preprocess(IExampleSet dataset) {
 
             if (attr.isNominal()) {
                 // get orders
-                Integer[] valuesOrder = new Integer[attr.getMapping().size()];
+                List<Integer> valuesOrder = new ArrayList<Integer>();
                 List<String> labels = new ArrayList<>();
                 labels.addAll(attr.getMapping().getValues());
                 Collections.sort(labels);
                 for (int j = 0; j < labels.size(); ++j) {
-                    valuesOrder[j] = attr.getMapping().getIndex(labels.get(j));
+                    int index = attr.getMapping().getIndex(labels.get(j));
+
+                    if (trainSet.getDoubleColumn(attr).contains((double)index)) {
+                        valuesOrder.add(index);
+                    }
                 }
                 attributeValuesOrder.put(attr, valuesOrder);
             }
@@ -431,8 +435,8 @@ class Stats {
                     stats[0] = new Stats(0, 0, 0);
                     stats[1] = new Stats(finalCovered_p - stats[0].p, finalCovered_n - stats[0].n, finalCovered_new_p - stats[0].p_new);
 
-                    for (int j = 0; j < attr.getMapping().size(); ++j) {
-                        int bid = attributeValuesOrder.get(attr)[j];
+                    for (int j = 0; j < attributeValuesOrder.get(attr).size(); ++j) {
+                        int bid = attributeValuesOrder.get(attr).get(j);
 
                         // update stats
                         stats[0].p = cur_positives[bid];

adaa.analytics.rules/src/main/java/adaa/analytics/rules/logic/induction/ClassificationFinder.java

@@ -729,13 +729,13 @@ class TotalPosNeg {
 			} else {
 				// unweighted case
 				// try all possible conditions
-				Integer[] attributeValueOrder = attributeValuesOrder.get(attr);
+				List<Integer> attributeValueOrder = attributeValuesOrder.get(attr);
 				Map<Double, IntegerBitSet> precalculatedCovering = precalculatedCoverings.get(attr);
 				Map<Double, IntegerBitSet> precalculatedCoveringComplement = precalculatedCoveringsComplement.get(attr);
-				for (int j = 0; j < attr.getMapping().size(); ++j) {
+				for (int j = 0; j < attributeValueOrder.size(); ++j) {
 
 					// evaluate straight condition
-					int i = attributeValueOrder[j];
+					int i = attributeValueOrder.get(j);
 					IntegerBitSet conditionCovered = precalculatedCovering.get((double) i);
 					double p = conditionCovered.calculateIntersectionSize(rule.getCoveredPositives());
 					int toCover_p = conditionCovered.calculateIntersectionSize((IntegerBitSet) coveredByRule, (IntegerBitSet) uncoveredPositives);

adaa.analytics.rules/src/main/java/adaa/analytics/rules/logic/induction/RegressionFinder.java

@@ -241,10 +241,10 @@ class Stats{
 
 				} else {
 					// try all possible conditions
-					for (int j = 0; j < attr.getMapping().size(); ++j) {
+					for (int j = 0; j < attributeValuesOrder.get(attr).size(); ++j) {
 
 						// evaluate straight condition
-						int i = attributeValuesOrder.get(attr)[j];
+						int i = attributeValuesOrder.get(attr).get(j);
 						ElementaryCondition candidate = new ElementaryCondition(
 								attr.getName(), new SingletonSet((double)i, attr.getMapping().getValues()));
 						checkCandidate(dataset, rule, candidate, uncovered, covered, best);
@@ -358,10 +358,10 @@ protected ElementaryCondition induceCondition(
 					}
 				} else {
 					// try all possible conditions
-					for (int j = 0; j < attr.getMapping().size(); ++j) {
+					for (int j = 0; j < attributeValuesOrder.get(attr).size(); ++j) {
 
 						// evaluate straight condition
-						int i = attributeValuesOrder.get(attr)[j];
+						int i = attributeValuesOrder.get(attr).get(j);
 						ElementaryCondition candidate = new ElementaryCondition(
 								attr.getName(), new SingletonSet((double)i, attr.getMapping().getValues()));
 						checkCandidate(dataset, rule, candidate, uncovered, covered, best);