iPFI feature importance

moa/src/main/java/moa/learners/featureanalysis/ClassifierWithFeatureImportance.java

@@ -25,6 +25,7 @@
 import com.yahoo.labs.samoa.instances.Instance;
 import moa.classifiers.AbstractClassifier;
 import moa.classifiers.MultiClassClassifier;
+import moa.classifiers.Regressor;
 import moa.classifiers.meta.AdaptiveRandomForest;
 import moa.core.Measurement;
 import moa.core.Utils;
@@ -64,7 +65,7 @@ public String getPurposeString() {
 
     public ClassOption featureImportanceLearnerOption = new ClassOption("featureImportanceLearner", 'l',
             "Learner used to build the model from which the feature importances are extracted",
-            FeatureImportanceClassifier.class, "moa.learners.featureanalysis.FeatureImportanceHoeffdingTree");
+            FeatureImportanceLearner.class, "moa.learners.featureanalysis.FeatureImportanceHoeffdingTree");
 
     public FlagOption doNotNormalizeFeatureScoreOption = new FlagOption("doNotNormalizeFeatureScore", 'n',
             "If set the feature importances will not be normalized");
@@ -84,7 +85,7 @@ public String getPurposeString() {
     protected PrintStream debugStream;
 
     protected long instancesSeen = 0;
-    protected FeatureImportanceClassifier featureImportanceClassifierLearner;
+    protected FeatureImportanceLearner featureImportanceClassifierLearner;
 
     protected double mean = -1.0;
     protected double median = -1.0;
@@ -113,7 +114,14 @@ protected void createDebugOutputFile() {
     public void resetLearningImpl() {
         this.instancesSeen = 0;
         this.featureImportanceClassifierLearner = null;
-        this.featureImportanceClassifierLearner = (FeatureImportanceClassifier) getPreparedClassOption(this.featureImportanceLearnerOption);
+        this.featureImportanceClassifierLearner = (FeatureImportanceLearner) getPreparedClassOption(this.featureImportanceLearnerOption);
+        // FeatureImportanceLearner covers regressors as well, so they reach the option chooser
+        // here even though this wrapper reads the votes as a class distribution.
+        if (this.featureImportanceClassifierLearner instanceof Regressor) {
+            throw new IllegalArgumentException(this.getClass().getName() + " needs a classifier, "
+                    + "but " + this.featureImportanceClassifierLearner.getClass().getName()
+                    + " is a regressor.");
+        }
         this.featureImportanceClassifierLearner.resetLearning();
         this.createDebugOutputFile();
     }

moa/src/main/java/moa/learners/featureanalysis/FeatureImportanceHoeffdingTree.java

@@ -32,7 +32,7 @@
  * @version $Revision: 1 $
  */
 public class FeatureImportanceHoeffdingTree extends AbstractClassifier implements MultiClassClassifier,
-        CapabilitiesHandler, FeatureImportanceClassifier {
+        CapabilitiesHandler, FeatureImportanceLearner {
 
     public ClassOption treeLearnerOption = new ClassOption("treeLearner", 'l',
             "Decision Tree learner.", HoeffdingTree.class,

moa/src/main/java/moa/learners/featureanalysis/FeatureImportanceHoeffdingTreeEnsemble.java

@@ -32,7 +32,7 @@
  * @version $Revision: 1 $
  */
 public class FeatureImportanceHoeffdingTreeEnsemble extends AbstractClassifier implements MultiClassClassifier,
-        CapabilitiesHandler, FeatureImportanceClassifier {
+        CapabilitiesHandler, FeatureImportanceLearner {
 
     public ClassOption ensembleLearnerOption = new ClassOption("ensembleLearner", 'l',
             "Ensemble learner to train and analyze.", Classifier.class,

moa/src/main/java/moa/learners/featureanalysis/FeatureImportanceClassifier.java → moa/src/main/java/moa/learners/featureanalysis/FeatureImportanceLearner.java

@@ -1,7 +1,6 @@
 /*
- *    FeatureScore.java
- *    Copyright (C) 2020 University of Waikato, Hamilton, New Zealand
- *    @author Heitor Murilo Gomes (hgomes at waikato dot ac dot nz)
+ *    FeatureImportanceLearner.java
+ *    Copyright (C) 2026 University of Waikato, Hamilton, New Zealand
  *
  *    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
@@ -22,34 +21,37 @@
 import moa.classifiers.Classifier;
 
 /**
- * Feature Importance Classifier
+ * Feature Importance Learner
  *
- * <p>This interface defines the methods to be implemented on a Classifier to allow it to produce feature importances.
- * </p>
+ * <p>This interface defines the methods to be implemented on a learner to allow it to produce
+ * feature importances. Nothing here depends on the kind of prediction being made, so it covers
+ * classifiers and regressors alike; a regressor additionally declares
+ * {@link moa.classifiers.Regressor} so that the regression tasks and the regression tab of the
+ * GUI pick it up.</p>
  *
- * <p>See details in:<br> Heitor Murilo Gomes, Rodrigo Fernandes de Mello, Bernhard Pfahringer, Albert Bifet.
- * Feature Scoring using Tree-Based Ensembles for Evolving Data Streams.
+ * <p>See details in:<br> Heitor Murilo Gomes, Rodrigo Fernandes de Mello, Bernhard Pfahringer,
+ * Albert Bifet. Feature Scoring using Tree-Based Ensembles for Evolving Data Streams.
  * IEEE International Conference on Big Data (pp. 761-769), 2019</p>
- * </p>
  *
  * @author Heitor Murilo Gomes
  */
-public interface FeatureImportanceClassifier extends Classifier {
+public interface FeatureImportanceLearner extends Classifier {
 
     /**
      * Obtain the current importance for each feature.
      *
+     * @param normalize whether to rescale the scores before returning them
      * @return array containing the importance/score estimated for each feature
      */
     double[] getFeatureImportances(boolean normalize);
 
     /**
-     * The output is a double array where values indicates the
-     * original feature index and the order of the array its
-     * ranking. The size of this array is expected to be less than
-     * the complete set of features.
-     * @param k
-     * @param normalize
+     * The output is an array where values indicate the original feature index and the order of
+     * the array its ranking. The size of this array is expected to be less than or equal to the
+     * complete set of features.
+     *
+     * @param k how many features to return
+     * @param normalize whether to rank the normalized scores
      * @return the k features with the highest scores.
      */
     int[] getTopKFeatures(int k, boolean normalize);