@Test
public void testClone() {
System.out.println("clone");
for (boolean useCatFeatures : new boolean[] {true, false}) {
RandomForest instance = new RandomForest();
ClassificationDataSet t1 = FixedProblems.getSimpleKClassLinear(100, 3);
ClassificationDataSet t2 = FixedProblems.getSimpleKClassLinear(100, 6);
if (useCatFeatures) {
t1.applyTransform(new NumericalToHistogram(t1));
t2.applyTransform(new NumericalToHistogram(t2));
}
instance = instance.clone();
instance.trainC(t1);
RandomForest result = instance.clone();
for (int i = 0; i < t1.getSampleSize(); i++)
assertEquals(t1.getDataPointCategory(i), result.classify(t1.getDataPoint(i)).mostLikely());
result.trainC(t2);
for (int i = 0; i < t1.getSampleSize(); i++)
assertEquals(
t1.getDataPointCategory(i), instance.classify(t1.getDataPoint(i)).mostLikely());
for (int i = 0; i < t2.getSampleSize(); i++)
assertEquals(t2.getDataPointCategory(i), result.classify(t2.getDataPoint(i)).mostLikely());
}
}
@Test
public void testTrainC_ClassificationDataSetMissingFeat() {
System.out.println("trainC");
for (boolean useCatFeatures : new boolean[] {true, false}) {
RandomForest instance = new RandomForest();
ClassificationDataSet train = FixedProblems.getCircles(1000, 1.0, 10.0, 100.0);
// RF may not get boundry perfect, so use noiseless for testing
ClassificationDataSet test =
FixedProblems.getCircles(1000, 0.0, new XORWOW(), 1.0, 10.0, 100.0);
train.applyTransform(new InsertMissingValuesTransform(0.1));
test.applyTransform(new InsertMissingValuesTransform(0.01));
ClassificationModelEvaluation cme = new ClassificationModelEvaluation(instance, train);
if (useCatFeatures)
cme.setDataTransformProcess(
new DataTransformProcess(
new NumericalToHistogram.NumericalToHistogramTransformFactory()));
cme.evaluateTestSet(test);
if (useCatFeatures) // hard to get right with only 2 features like this
assertTrue(cme.getErrorRate() <= 0.17);
else assertTrue(cme.getErrorRate() <= 0.1);
}
}
@Test
public void testTrainC_ClassificationDataSet_ExecutorService() {
System.out.println("trainC");
for (boolean useCatFeatures : new boolean[] {true, false}) {
RandomForest instance = new RandomForest();
ExecutorService ex = Executors.newFixedThreadPool(SystemInfo.LogicalCores);
ClassificationDataSet train = FixedProblems.getCircles(1000, 1.0, 10.0, 100.0);
// RF may not get boundry perfect, so use noiseless for testing
ClassificationDataSet test =
FixedProblems.getCircles(100, 0.0, new XORWOW(), 1.0, 10.0, 100.0);
ClassificationModelEvaluation cme = new ClassificationModelEvaluation(instance, train, ex);
if (useCatFeatures)
cme.setDataTransformProcess(
new DataTransformProcess(
new NumericalToHistogram.NumericalToHistogramTransformFactory()));
cme.evaluateTestSet(test);
assertTrue(cme.getErrorRate() <= 0.001);
ex.shutdownNow();
}
}
@Test
public void testTrainC_ClassificationDataSet() {
System.out.println("trainC");
for (boolean conc : new boolean[] {true, false}) {
OneVSOne instance = new OneVSOne(new DCDs(), conc);
ClassificationDataSet train = FixedProblems.getSimpleKClassLinear(1000, 7);
ClassificationDataSet test = FixedProblems.getSimpleKClassLinear(100, 7);
ClassificationModelEvaluation cme = new ClassificationModelEvaluation(instance, train);
cme.evaluateTestSet(test);
assertTrue(cme.getErrorRate() <= 0.001);
}
}
@Test
public void testTrainC_ClassificationDataSet_ExecutorService() {
System.out.println("trainC");
for (boolean conc : new boolean[] {true, false}) {
OneVSOne instance = new OneVSOne(new DCDs(), conc);
ClassificationDataSet train = FixedProblems.getSimpleKClassLinear(1000, 7);
ClassificationDataSet test = FixedProblems.getSimpleKClassLinear(100, 7);
ExecutorService ex = Executors.newFixedThreadPool(SystemInfo.LogicalCores);
ClassificationModelEvaluation cme = new ClassificationModelEvaluation(instance, train, ex);
cme.evaluateTestSet(test);
assertTrue(cme.getErrorRate() <= 0.001);
}
}
@Test
public void testTrainC_RegressionDataSet() {
System.out.println("train");
for (boolean useCatFeatures : new boolean[] {true, false}) {
RandomForest instance = new RandomForest();
RegressionDataSet train = FixedProblems.getLinearRegression(1000, new XORWOW(), coefs);
RegressionDataSet test = FixedProblems.getLinearRegression(100, new XORWOW(), coefs);
RegressionModelEvaluation rme = new RegressionModelEvaluation(instance, train);
if (useCatFeatures)
rme.setDataTransformProcess(
new DataTransformProcess(
new NumericalToHistogram.NumericalToHistogramTransformFactory()));
rme.evaluateTestSet(test);
assertTrue(rme.getMeanError() <= test.getTargetValues().mean() * 2.5);
}
}
@Test
public void testClone() {
System.out.println("clone");
ClassificationDataSet t1 = FixedProblems.getSimpleKClassLinear(1000, 7);
ClassificationDataSet t2 = FixedProblems.getSimpleKClassLinear(1000, 9);
OneVSOne instance = new OneVSOne(new DCDs());
instance = instance.clone();
instance.trainC(t1);
OneVSOne result = instance.clone();
for (int i = 0; i < t1.getSampleSize(); i++)
assertEquals(t1.getDataPointCategory(i), result.classify(t1.getDataPoint(i)).mostLikely());
result.trainC(t2);
for (int i = 0; i < t1.getSampleSize(); i++)
assertEquals(t1.getDataPointCategory(i), instance.classify(t1.getDataPoint(i)).mostLikely());
for (int i = 0; i < t2.getSampleSize(); i++)
assertEquals(t2.getDataPointCategory(i), result.classify(t2.getDataPoint(i)).mostLikely());
}
