/*
* A redesigned Fitness Function calculator
* It takes into account the accuracy of the decision tree while classifying both, training and test examples
* The fitness score is a function of the weighted average of the two accuracies.
*/
private void calculateFitnessScore(double trainingWeight, double testingWeight)
throws OptimalScoreException {
DecisionTreeClassifier dtClassifier = getDecisionTree();
dtClassifier.TestAndFindAccuracy();
// Part 1 - Get training set accuracy
double trainingSetAccuracy = dtClassifier.getAccuracy();
// Part 2 - Get test set accuracy
SampleCollection test_samples =
new SampleCollection(
DataHolder.getTestingSamplesFileName(), DataHolder.getAttributesFileName());
// test_samples.discretizeSamples(Constants.DiscretizerAlgorithms.EQUAL_BINNING);
test_samples.discretizeSamplesBasedOnOtherSampleCollection(dtClassifier.getTrainingSamples());
dtClassifier.setTestingSamples(test_samples);
dtClassifier.TestAndFindAccuracy();
double testSetAccuracy = dtClassifier.getAccuracy();
fitnessScore =
(trainingWeight * trainingSetAccuracy + testingWeight * testSetAccuracy)
/ (trainingWeight + testingWeight);
// fitnessScore = trainingSetAccuracy; It was running very slowly that's why all this circus.
// We'll find a solution.
// fitnessScore = trainingSetAccuracy > testSetAccuracy ? trainingSetAccuracy : testSetAccuracy;
}
/*
* Returns a new decision tree that was created by using only the features present in the chromosome.
*/
public DecisionTreeClassifier getDecisionTree() {
ArrayList<String> features = new ArrayList<String>();
for (int i = 0; i < chromosome.length(); ++i)
if (chromosome.charAt(i) == '1') features.add(FeatureSuperSet.get(i));
DecisionTreeClassifier dtClassifier = new DecisionTreeClassifier(samples, features);
SampleCollection training_samples =
new SampleCollection(
samples.getSamplesFilename(Filenames.TRAINING_SAMPLES_FILE),
samples.getSamplesFilename(Filenames.FEATURES_FILE));
training_samples.discretizeSamples(Constants.DiscretizerAlgorithms.EQUAL_BINNING);
dtClassifier.setTestingSamples(training_samples);
dtClassifier.setTrainingSamples(training_samples);
return dtClassifier;
}