Refactored performance measure after discretization changes #21

AutoTuning/src/main/java/LossFunctions/PredictionModel.java

@@ -0,0 +1,81 @@
+package LossFunctions;
+
+import de.viadee.xai.anchor.adapter.tabular.TabularInstance;
+import de.viadee.xai.anchor.algorithm.AnchorResult;
+
+import java.util.ArrayList;
+import java.util.Comparator;
+import java.util.List;
+
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+public class PredictionModel {
+
+    private static final Logger LOGGER = LoggerFactory.getLogger(PredictionModel.class);
+
+    /**
+     * - if one rule applies predict the label according to rule
+     * - if two rules apply predict according to more precise rule
+     * - if both rules are equally precise predict randomly
+     */
+
+    private final List<AnchorResult<TabularInstance>> rules;
+
+    public PredictionModel(List<AnchorResult<TabularInstance>> globalExplanations) {
+        this.rules = globalExplanations;
+        // sort by precision so higher precisions get prioritized
+        rules.sort(Comparator.comparingDouble(AnchorResult<TabularInstance>::getPrecision).reversed());
+
+    }
+
+    /**
+     * @param instances
+     * @return
+     */
+    public List<Integer> predict(TabularInstance[] instances) {
+
+        List<Integer> predictions = new ArrayList<>();
+
+        for (TabularInstance instance : instances) {
+            predictions.add(this.predictSingle(instance));
+        }
+        return predictions;
+    }
+
+    /**
+     * @param instance
+     * @return
+     */
+    public int predictSingle(TabularInstance instance) {
+
+        int ruleNumber = 0;
+
+        for (AnchorResult<TabularInstance> rule : this.rules) {
+            int numberMatches = 0;
+            ruleNumber++;
+//            System.out.println("Check rule " + ruleNumber + " with precision " + r.getPrecision() + " and label " + r.getLabel());
+
+            for (int f : rule.getCanonicalFeatures()) {
+
+                double instanceValue = instance.getValue(f);
+                double ruleValue = rule.getInstance().getValue(f);
+
+                System.out.println("Feature: " + rule.getInstance().getFeatures()[f].getName() + " - Instance value: " + instanceValue + " ---- Rule value: " + ruleValue);
+
+                if (instanceValue == ruleValue) {
+                    numberMatches++;
+                }
+            }
+
+            if (numberMatches == (rule.getCanonicalFeatures().size())) {
+                LOGGER.info("Predict " + rule.getLabel() + " for the instance based on rule " + ruleNumber + ".");
+                return rule.getLabel();
+            }
+        }
+
+        LOGGER.info("No rule found to predict the instance.");
+        return -1;
+    }
+
+}

AutoTuning/src/main/java/RandomSearch/RandomSearch.java

@@ -1,6 +1,6 @@
 package RandomSearch;
 
-import LossFunctions.Accuracy.PredictionModel;
+import LossFunctions.PredictionModel;
 import LossFunctions.PerformanceMeasures;
 import de.viadee.xai.anchor.adapter.tabular.AnchorTabular;
 import de.viadee.xai.anchor.adapter.tabular.TabularInstance;