/**
* Tests a classifier on a data set
*
* @param cls the classifier to test
* @param data the data set to test on
* @return the performance for each class
*/
public static Map<Object, PerformanceMeasure> testDataset(Classifier cls, Dataset data) {
Map<Object, PerformanceMeasure> out = new HashMap<Object, PerformanceMeasure>();
for (Object o : data.classes()) {
out.put(o, new PerformanceMeasure());
}
for (Instance instance : data) {
Object prediction = cls.classify(instance);
if (instance.classValue().equals(prediction)) { // prediction
// ==class
for (Object o : out.keySet()) {
if (o.equals(instance.classValue())) {
out.get(o).tp++;
} else {
out.get(o).tn++;
}
}
} else { // prediction != class
for (Object o : out.keySet()) {
/* prediction is positive class */
if (prediction.equals(o)) {
out.get(o).fp++;
}
/* instance is positive class */
else if (o.equals(instance.classValue())) {
out.get(o).fn++;
}
/* none is positive class */
else {
out.get(o).tn++;
}
}
}
}
return out;
}
/** Shows the default usage of the KNN algorithm. */
public static void main(String[] args) throws Exception {
/* Load a data set */
Dataset data = FileHandler.loadDataset(new File(DATASET), 4, ",");
/*
* Contruct a KNN classifier that uses 5 neighbors to make a decision.
*/
Classifier knn = new KNearestNeighbors(5);
knn.buildClassifier(data);
Classifier kdtKnn = new KDtreeKNN(5);
kdtKnn.buildClassifier(data);
/*
* Load a data set for evaluation, this can be a different one, but we
* will use the same one.
*/
Dataset dataForClassification = FileHandler.loadDataset(new File(DATASET), 4, ",");
/* Counters for correct and wrong predictions. */
int correct = 0, wrong = 0;
/* Classify all instances and check with the correct class values */
for (Instance inst : dataForClassification) {
Object predictedClassValue = knn.classify(inst);
Object realClassValue = inst.classValue();
// System.out.println("predicted=" + predictedClassValue.toString() + "
// real="+realClassValue.toString());
if (predictedClassValue.equals(realClassValue)) {
correct++;
} else {
wrong++;
}
}
System.out.println("Correct predictions " + correct);
System.out.println("Wrong predictions " + wrong);
/* Performance
*
*/
System.out.println("Performance ...");
Map<Object, PerformanceMeasure> pm = EvaluateDataset.testDataset(knn, dataForClassification);
printPerfMeasure(pm);
/*
* Cross validation
*/
System.out.println("Cross validation ...");
/* Construct new cross validation instance with the KNN classifier, */
CrossValidation cv = new CrossValidation(knn);
/* 5-fold CV with fixed random generator */
Map<Object, PerformanceMeasure> p0 = cv.crossValidation(data, 5, new Random(1));
Map<Object, PerformanceMeasure> p1 = cv.crossValidation(data, 5, new Random(1));
Map<Object, PerformanceMeasure> p2 = cv.crossValidation(data, 5, new Random(25));
printPerfMeasure(p0);
printPerfMeasure(p1);
printPerfMeasure(p2);
/*
* Create Weka classifier
*/
System.out.println("Weka classifier ...");
SMO smo = new SMO();
/* Wrap Weka classifier in bridge */
Classifier javamlsmo = new WekaClassifier(smo);
/* Initialize cross-validation */
CrossValidation wekaCV = new CrossValidation(javamlsmo);
/* Perform cross-validation */
Map<Object, PerformanceMeasure> wekaPm = wekaCV.crossValidation(data);
/* Output results
* see http://en.wikipedia.org/wiki/Precision_and_recall
*/
System.out.println("see http://en.wikipedia.org/wiki/Precision_and_recall" + wekaPm);
printPerfMeasure(wekaPm);
/*
* Feature scoring
*/
System.out.println("Feature scoring ");
GainRatio ga = new GainRatio();
/* Apply the algorithm to the data set */
ga.build(data);
/* Print out the score of each attribute */
for (int i = 0; i < ga.noAttributes(); i++) {
System.out.println("Attribute[" + i + "] relevance" + ga.score(i));
}
}