@Test
public void testFactorNACalc() {
FactorExtraTrees et = getFET(10, 5, false);
FactorExtraTrees etw = getFET(10, 5, true);
double gini;
gini = 1 - (0.4 * 0.4 + 0.6 * 0.6);
assertEquals(gini, et.get1NaNScore(AbstractTrees.seq(10)), 1e-6);
gini = 1 - (Math.pow(3.0 / 9.0, 2) + Math.pow(6.0 / 9.0, 2));
assertEquals(gini, et.get1NaNScore(AbstractTrees.seq(9)), 1e-6);
// testing NaN counts:
CutResult cr;
cr = new CutResult();
et.calculateCutScore(AbstractTrees.seq(9), 2, 0.5, cr);
assertEquals(3.0, cr.nanWeigth, 1e-6);
cr = new CutResult();
et.calculateCutScore(AbstractTrees.seq(9), 1, 0.5, cr);
assertEquals(0.0, cr.nanWeigth, 1e-6);
// testing weights:
cr = new CutResult();
etw.calculateCutScore(AbstractTrees.seq(9), 2, 0.5, cr);
assertEquals(1.5, cr.nanWeigth, 1e-6);
}
/**
* @param ndata
* @param ndim
*/
public static FactorExtraTrees getFET(int ndata, int ndim, boolean useWeights) {
int[] output = new int[ndata];
Matrix m = new Matrix(ndata, ndim);
// generate values for all outputs
for (int row = 0; row < output.length; row++) {
m.set(row, 1, row / (double) output.length);
m.set(row, 2, 0.5);
if (row == 5 || row == 6 || row == 7) {
m.set(row, 2, Double.NaN);
}
output[row] = m.get(row, 1) > 0.55 ? 1 : 0;
}
FactorExtraTrees et = new FactorExtraTrees(m, output);
et.setHasNaN(true);
if (useWeights) {
double[] w = new double[ndata];
for (int i = 0; i < w.length; i++) {
w[i] = 0.5;
}
et.setWeights(w);
}
return et;
}
@Test
public void testFactorNALearn() {
int ndim = 5;
FactorExtraTrees et = getFET(100, ndim, false);
FactorExtraTrees etw = getFET(100, ndim, true);
et.learnTrees(3, 3, 5);
etw.learnTrees(3, 3, 5);
double[] x = new double[ndim];
for (int i = 0; i < x.length; i++) {
x[i] = Double.NaN;
}
int[] val;
val = et.getValues(new Matrix(x, 1, ndim));
assertEquals(-1, val[0]);
val = etw.getValues(new Matrix(x, 1, ndim));
assertEquals(-1, val[0]);
val = et.getValues(et.input);
for (int i = 0; i < val.length; i++) {
assertTrue(val[i] >= 0);
}
}
