private Mx getDropoutMask() {
final Mx dropoutMask = new VecBasedMx(output.rows(), output.columns());
for (int i = 0; i < dropoutMask.dim(); i++) {
dropoutMask.set(i, Math.random() > dropoutFraction ? 1 : 0);
}
return dropoutMask;
}
@Override
public ExponentialObliviousTree fit(final VecDataSet ds, final L2 loss) {
final ObliviousTree base = got.fit(ds, loss);
features = base.features();
double baseMse = 0;
for (int i = 0; i < ds.length(); i++)
baseMse += sqr(base.value(ds.data().row(i)) - loss.target.get(i));
System.out.println("\nBase_MSE = " + baseMse);
if (features.size() != depth) {
System.out.println("Oblivious Tree bug");
try {
final PrintWriter printWriter = new PrintWriter(new File("badloss.txt"));
for (int i = 0; i < ds.length(); i++) printWriter.println(loss.target.get(i));
printWriter.close();
} catch (FileNotFoundException e) {
e.printStackTrace();
}
System.exit(-1);
}
precalculateMissCoefficients(ds, loss);
// System.out.println("Precalc is over");
final double[][] out = new double[1 << depth][(depth + 1) * (depth + 2) / 2];
for (int index = 0; index < 1 << depth; index++) {
final Mx a = new VecBasedMx(numberOfVariablesByLeaf, numberOfVariablesByLeaf);
final Vec b = new ArrayVec(numberOfVariablesByLeaf);
for (int i = 0; i < numberOfVariablesByLeaf; i++) b.set(i, -linearMissCoefficient[index][i]);
for (int i = 0; i < numberOfVariablesByLeaf; i++)
for (int j = 0; j < numberOfVariablesByLeaf; j++)
a.set(i, j, quadraticMissCoefficient[index][i][j]);
for (int i = 0; i < numberOfVariablesByLeaf; i++) a.adjust(i, i, 1e-1);
final Vec value = GreedyPolynomialExponentRegion.solveLinearEquationUsingLQ(a, b);
// System.out.println(a);
for (int k = 0; k <= depth; k++)
for (int j = 0; j <= k; j++) out[index][k * (k + 1) / 2 + j] = value.get(getIndex(0, k, j));
/*if(quadraticMissCoefficient[index][0][0] != 0)
out[index][0] = linearMissCoefficient[index][0] / quadraticMissCoefficient[index][0][0];*/
// out[index][0] = base.values()[index];
// for (int i = 0; i < out[index].length; i++)
// System.out.println(out[index][i]);
}
// for(int i =0 ; i < gradLambdas.size();i++)
// System.out.println(serializeCondtion(i));
final ExponentialObliviousTree ret = new ExponentialObliviousTree(features, out, DistCoef);
double mse = 0;
for (int i = 0; i < ds.length(); i++)
mse += sqr(ret.value(ds.data().row(i)) - loss.target.get(i));
System.out.println("MSE = " + mse);
/*if (mse > baseMse + 1e-5)
try {
throw new Exception("Bad model work mse of based model less than mse of extended model");
} catch (Exception e) {
e.printStackTrace();
//System.exit(-1);
}*/
return ret;
}
private Mx leftContract(final Mx original) {
final VecBasedMx contracted = new VecBasedMx(original.rows(), original.columns() - 1);
for (int i = 0; i < contracted.rows(); i++) {
for (int j = 0; j < contracted.columns(); j++) {
contracted.set(i, j, original.get(i, j + 1));
}
}
return contracted;
}
@Override
public double value(final Vec x) {
final Mx predictMx = (Mx) x;
int count = 0;
for (int i = 0; i < predictMx.rows(); i++) {
if (VecTools.distance(predictMx.row(i), targets.row(i)) < MathTools.EPSILON) {
count++;
}
}
return (double) count / targets.rows();
}
public CMLMetricOptimization(
final VecDataSet ds,
final BlockwiseMLLLogit target,
final Mx S,
final double c,
final double step) {
this.ds = ds;
this.target = target;
this.step = step;
this.classesIdxs = MCTools.splitClassesIdxs(target.labels());
this.laplacian = VecTools.copy(S);
VecTools.scale(laplacian, -1.0);
for (int i = 0; i < laplacian.rows(); i++) {
final double diagElem = VecTools.sum(S.row(i));
laplacian.adjust(i, i, diagElem);
}
this.c = c;
}
public Mx trainProbs(final Mx codingMatrix, final Func[] binClassifiers) {
final Mx result = new VecBasedMx(codingMatrix.rows(), codingMatrix.columns());
for (int l = 0; l < result.columns(); l++) {
System.out.println("Optimize column " + l);
final FuncC1 columnTargetFunction = new ColumnTargetFunction(binClassifiers[l]);
final Vec muColumn = optimizeColumn(columnTargetFunction, codingMatrix.col(l));
VecTools.assign(result.col(l), muColumn);
}
return result;
}
private Mx leftExtend(final Mx original) {
final VecBasedMx extended = new VecBasedMx(original.rows(), original.columns() + 1);
for (int i = 0; i < original.rows(); i++) {
extended.set(i, 0, bias);
}
for (int i = 0; i < original.rows(); i++) {
for (int j = 1; j < original.columns() + 1; j++) {
extended.set(i, j, original.get(i, j - 1));
}
}
return extended;
}
public void forward() {
if (bias != 0) {
activations = leftExtend(input);
} else {
activations = VecTools.copy(input);
}
output = MxTools.multiply(activations, MxTools.transpose(weights));
rectifier.value(output, output);
if (dropoutFraction > 0) {
if (isTrain) {
dropoutMask = getDropoutMask();
for (int i = 0; i < output.dim(); i++) {
output.set(i, output.get(i) * dropoutMask.get(i));
}
} else {
for (int i = 0; i < output.dim(); i++) {
output.set(i, output.get(i) * (1 - dropoutFraction));
}
}
}
}
@Override
public int dim() {
return targets.dim();
}
public void backward() {
Mx cnc = null;
if (bias_b != 0) {
cnc = leftContract(output);
} else {
cnc = VecTools.copy(output);
}
difference = MxTools.multiply(MxTools.transpose(cnc), activations);
for (int i = 0; i < difference.dim(); i++) {
difference.set(i, difference.get(i) / activations.rows());
}
input = MxTools.multiply(cnc, weights);
rectifier.grad(activations, activations);
for (int i = 0; i < input.dim(); i++) {
input.set(i, input.get(i) * activations.get(i));
if (dropoutFraction > 0) {
input.set(i, input.get(i) * dropoutMask.get(i));
}
}
}
