Print precision and recall to logs

Author: oshando

swan-pipeline/src/main/java/de/fraunhofer/iem/swan/model/MonteCarloValidator.java

@@ -19,11 +19,15 @@ public class MonteCarloValidator {
 
     private ArrayList<String> predictions;
     private HashMap<String, ArrayList<Double>> fMeasure;
+    private HashMap<String, ArrayList<Double>> precision;
+    private HashMap<String, ArrayList<Double>> recall;
     private DecimalFormat df = new DecimalFormat("####0.00");
 
     public MonteCarloValidator() {
         predictions = new ArrayList<>();
         fMeasure = new HashMap<>();
+        precision = new HashMap<>();
+        recall = new HashMap<>();
     }
 
     public ArrayList<String> getPredictions() {
@@ -34,6 +38,14 @@ public HashMap<String, ArrayList<Double>> getFMeasure() {
         return fMeasure;
     }
 
+    public HashMap<String, ArrayList<Double>> getRecall() {
+        return recall;
+    }
+
+    public HashMap<String, ArrayList<Double>> getPrecision() {
+        return precision;
+    }
+
     /**
      * Evaluates instances using Monte Carlo Cross Evaluation.
      *
@@ -55,11 +67,11 @@ public void monteCarloValidate(Instances instances, Classifier classifier, doubl
             Instances trainInstances = new Instances(instances, 0, trainSize);
             Instances testInstances = new Instances(instances, trainSize, testSize);
 
-            evaluate(classifier, trainInstances, testInstances);
+            evaluate(classifier, trainInstances, testInstances, i);
         }
     }
 
-    public ArrayList<String> evaluate(Classifier classifier, Instances trainInstances, Instances testInstances) {
+    public ArrayList<String> evaluate(Classifier classifier, Instances trainInstances, Instances testInstances, int iteration) {
 
         Evaluation eval = null;
         try {
@@ -75,6 +87,9 @@ public ArrayList<String> evaluate(Classifier classifier, Instances trainInstance
 
             eval.evaluateModel(classifier, testInstances, abstractOutput);
 
+            String header = "=== " + classifier.getClass().getSimpleName() + " Iteration #" + iteration + ": " + trainInstances.attribute(trainInstances.numAttributes() - 1).name() + " ===";
+            System.out.println(eval.toClassDetailsString(header));
+
             //Obtain all predictions and extract method signatures
             String[] output = abstractOutput.getBuffer().toString().split("\n");
 
@@ -99,10 +114,14 @@ public void updateResultSet(Instances instances, Evaluation eval) {
             String currentClass = instances.classAttribute().value(c);
 
             if (!currentClass.contentEquals("0")) {
-                if (!fMeasure.containsKey(currentClass))
+                if (!fMeasure.containsKey(currentClass)) {
                     fMeasure.put(currentClass, new ArrayList<>(Collections.singletonList(Double.isNaN(eval.fMeasure(c)) ? 0 : Double.parseDouble(df.format(eval.fMeasure(c))))));
-                else {
+                    recall.put(currentClass, new ArrayList<>(Collections.singletonList(Double.isNaN(eval.recall(c)) ? 0 : Double.parseDouble(df.format(eval.recall(c))))));
+                    precision.put(currentClass, new ArrayList<>(Collections.singletonList(Double.isNaN(eval.precision(c)) ? 0 : Double.parseDouble(df.format(eval.precision(c))))));
+                } else {
                     fMeasure.get(currentClass).add(Double.isNaN(eval.fMeasure(c)) ? 0 : Double.parseDouble(df.format(eval.fMeasure(c))));
+                    recall.get(currentClass).add(Double.isNaN(eval.recall(c)) ? 0 : Double.parseDouble(df.format(eval.recall(c))));
+                    precision.get(currentClass).add(Double.isNaN(eval.precision(c)) ? 0 : Double.parseDouble(df.format(eval.precision(c))));
                 }
             }
         }

swan-pipeline/src/main/java/de/fraunhofer/iem/swan/model/toolkit/Weka.java

@@ -42,7 +42,7 @@ public class Weka {
     private static final Logger logger = LoggerFactory.getLogger(Weka.class);
     private HashMap<String, ArrayList<Category>> predictions;
     private HashMap<String, HashMap<String, ArrayList<Double>>> results;
-    private DecimalFormat df = new DecimalFormat("####0.00");
+    private DecimalFormat df = new DecimalFormat("####0.000");
 
     public Weka(WekaFeatureSet features, SwanOptions options, Set<Method> methods) {
         this.features = features;
@@ -164,6 +164,9 @@ public Pair<String, Double> crossValidateModel(Instances instances) {
             for (String key : evaluator.getFMeasure().keySet()) {
 
                 double averageFMeasure = evaluator.getFMeasure().get(key).stream().mapToDouble(a -> a).average().getAsDouble();
+                double averagePrecision = evaluator.getPrecision().get(key).stream().mapToDouble(a -> a).average().getAsDouble();
+                double averageRecall = evaluator.getRecall().get(key).stream().mapToDouble(a -> a).average().getAsDouble();
+
                 Pair summary = new Pair<>(classifier.getClass().getSimpleName(), Double.parseDouble(df.format(averageFMeasure)));
                 classifierSummary.add(summary);
 
@@ -183,8 +186,7 @@ public Pair<String, Double> crossValidateModel(Instances instances) {
                 } else
                     measure.put(classifier.getClass().getSimpleName(), evaluator.getFMeasure().get(key));
 
-                logger.debug("Average F-measure for {}({}) using {}: {}, {}", category, key, classifier.getClass().getSimpleName(),
-                        averageFMeasure, evaluator.getFMeasure().get(key));
+                logger.debug("{} Average F-measure ({}), Precision ({}) and Recall ({}) for {}({}) ", classifier.getClass().getSimpleName(), averageFMeasure,averagePrecision, averageRecall, category, key);
             }
             if (!category.contains("authentication"))
                 results.put(category, measure);