Precision and Recall added

ROC output renamed

build.properties

@@ -1,3 +1,3 @@
 extension.version=2
 extension.revision=0
-extension.update=002
+extension.update=003

resources/com/rapidminer/resources/i18n/OperatorsDocAnomalyDetection.xml

@@ -209,6 +209,9 @@
 			guesses the two classes (normal or outlier), a ROC of 0.5 would be
 			the result on average. As input, this operator expects the outlier score
 			as well as the ground truth as labels.
+			Additionally the operator returns the precision and recall values for the whole input data.
+			The precision is calculated as true_positives/(true_positives+false_positives).
+			The recall is calculated as true_positives/(number of all outliers).
 			</p>
 		</help>
 	</operator>
@@ -280,4 +283,4 @@
 		    </p>
 		</help>
 	</operator>	
-</operatorHelp>
+</operatorHelp>

src/de/dfki/madm/anomalydetection/evaluator/evaluation/ROCEvaluator.java

@@ -57,7 +57,7 @@ public String toString() {
 	public String getNormalClass() {
 		return this.normal;
 	}
-
+	public Object[][] pre = null; //prediction / recall
 	/**
 	 * The returned array has 4 columns denoting: false positive rate, 
 	 * true positive rate/recall, precision and the decision threshold   
@@ -113,7 +113,7 @@ public Object[][] evaluate(String outlierString, Object[] labels, double[] res)
 			if (j != size - 1 && outliers[j].outlierScore == outliers[j + 1].outlierScore)
 				continue;
 			Area += last[1] * ((double)falsePositive - last[0]) + (double)0.5 * ((double)falsePositive - last[0]) * ((double)truePositive - last[1]);
-			rocPoints.add(new double[] { falsePositive, truePositive , truePositive*1.0/(positive+negative), outliers[j].outlierScore});
+			rocPoints.add(new double[] { falsePositive, truePositive , truePositive*1.0/(truePositive+falsePositive), outliers[j].outlierScore});
 			last[0] = falsePositive;
 			last[1] = truePositive;
 
@@ -127,13 +127,17 @@ public Object[][] evaluate(String outlierString, Object[] labels, double[] res)
 		double totalArea = (double)positive * (double)negative;
 
 		auc = Area / totalArea;
-		result = new Object[rocPoints.size()][4];
+		result = new Object[rocPoints.size()][2];
 		int i = 0;
+		pre = new Object[rocPoints.size()][2];
 		for (double[] r : rocPoints) {
 			result[i][0] = r[0] / negative;
-			result[i][1] = r[1] / positive;
-			result[i][2]= r[2];
-			result[i++][3]= r[3];
+			result[i++][1] = r[1] / positive;
+			}
+		i=0;
+		for(double[] r : rocPoints) {
+			pre[i][0] = r[2]; //precision = tp /(tp+fp)
+			pre[i++][1] = r[1] / positive; //recall = tp(so far) / all outlier
 		}
 
 		return result;

src/de/dfki/madm/anomalydetection/operator/evaluation/ROCOperator.java

@@ -55,6 +55,7 @@ public class ROCOperator extends Operator {
 	private OutputPort exampleSetOutput = getOutputPorts().createPort("example set");
 	private OutputPort rocExampleSet = getOutputPorts().createPort("roc set");
 	private OutputPort aucOutput = getOutputPorts().createPort("auc");
+	private OutputPort preOutput = getOutputPorts().createPort("pre");
 
 	private InputPort performanceInput = getInputPorts().createPort("performance");
 	private OutputPort performanceOutput = getOutputPorts().createPort("performance");
@@ -156,9 +157,14 @@ public void doWork() throws OperatorException {
 		ROCEvaluator evaluator = new ROCEvaluator();
 		String outlierString = getParameterAsString(PARAMETER_LABEL);
 		ExampleSet roc = (ExampleSet) ExampleSetFactory.createExampleSet(evaluator.evaluate(outlierString, labels, outliers));
+		ExampleSet pre = (ExampleSet) ExampleSetFactory.createExampleSet(evaluator.pre);
 		String norm = evaluator.getNormalClass();
-
+		roc.getAttributes().get("att1").setName("true_positive_rate");
+		roc.getAttributes().get("att2").setName("false_positive_rate");
+		pre.getAttributes().get("att1").setName("precision");
+		pre.getAttributes().get("att2").setName("recall");
 		rocExampleSet.deliver(roc);
+		preOutput.deliver(pre);
 		auc = evaluator.auc;
 		Object[][] auc_ = { { auc } };
 		Object[] labels2 = { "AUC" };