@Override
public DoubleMatrix2D transitionMatrix(double from_time, double to_time) {
double from_time_reminder = from_time % 1.0;
double from_time_div = from_time - from_time_reminder;
double to_time_reminder = to_time - from_time_div;
DoubleMatrix2D cached =
cachedTransitionMatrices.get(
new Pair<Double, Double>(from_time_reminder, to_time_reminder));
if (cached != null) {
return cached;
} else {
double step_start_time = from_time;
double step_end_time = step_start_time;
DoubleMatrix2D result = F.identity(num_states);
while (step_start_time < to_time) {
double step_start_time_reminder = step_start_time % 1.0;
double step_start_time_div = step_start_time - step_start_time_reminder;
if (isInSeason1(step_start_time)) {
step_end_time =
Math.min(
to_time,
step_start_time_div + season1Start + season1Length + infitesimalTimeInterval);
result =
result.zMult(
season1MigrationModel.transitionMatrix(step_start_time, step_end_time), null);
} else { // In Season 2
if (step_start_time_reminder < season1Start) {
step_end_time =
Math.min(to_time, step_start_time_div + season1Start + infitesimalTimeInterval);
} else {
step_end_time =
Math.min(
to_time, step_start_time_div + 1.0 + season1Start + infitesimalTimeInterval);
}
result =
result.zMult(
season2MigrationModel.transitionMatrix(step_start_time, step_end_time), null);
}
step_start_time = step_end_time;
}
// cache result
if (cachedTransitionMatrices.size() >= maxCachedTransitionMatrices) {
for (int i = 0; i < cachedTransitionMatrices.size() / 2; i++) {
cachedTransitionMatrices.remove(cachedTransitionMatrices.keySet().iterator().next());
}
}
cachedTransitionMatrices.put(
new Pair<Double, Double>(from_time_reminder, to_time_reminder), result);
return result;
}
}
@Test
public void testInclinedPlane() throws IOException {
DoubleMatrix1D normal = new DenseDoubleMatrix1D(3);
normal.assign(new double[] {.0, .0, 1.0});
InclinedPlane3D inclinedPlane = new InclinedPlane3D();
inclinedPlane.setRandomGenerator(new MersenneTwister(123456789));
inclinedPlane.setNormal(normal);
inclinedPlane.setBounds(new Rectangle(-5, -5, 10, 10));
inclinedPlane.setNoiseStd(0.5);
DoubleMatrix2D data = inclinedPlane.generate(10);
SVDPCA pca = new SVDPCA(data);
System.out.println("Eigenvalues:");
System.out.println(pca.getEigenvalues());
System.out.println("Eigenvectors:");
System.out.println(pca.getEigenvectors());
System.out.println("Meanvector:");
System.out.println(pca.getMean());
// Recalculate the input from a truncated SVD, first calculate the mean
DoubleMatrix1D mean = new SparseDoubleMatrix1D(3);
for (int i = 0; i < data.rows(); ++i) {
mean.assign(data.viewRow(i), Functions.plus);
}
mean.assign(Functions.div(data.rows()));
// Truncate the SVD and calculate the coefficient matrix
DenseDoubleMatrix2D coefficients = new DenseDoubleMatrix2D(data.rows(), 2);
DoubleMatrix2D centeredInput = data.copy();
for (int i = 0; i < data.rows(); ++i) {
centeredInput.viewRow(i).assign(mean, Functions.minus);
}
centeredInput.zMult(
pca.getEigenvectors().viewPart(0, 0, 2, 3), coefficients, 1, 0, false, true);
// Reconstruct the data from the lower dimensional information
DoubleMatrix2D reconstruction = data.copy();
for (int i = 0; i < reconstruction.rows(); ++i) {
reconstruction.viewRow(i).assign(mean);
}
coefficients.zMult(
pca.getEigenvectors().viewPart(0, 0, 2, 3), reconstruction, 1, 1, false, false);
// Output to file (can be read by GNU Plot)
String fileName = "inclined-plane-svd-pca.dat";
String packagePath = this.getClass().getPackage().getName().replaceAll("\\.", "/");
File outputFile = new File("src/test/resources/" + packagePath + "/" + fileName);
PrintWriter writer = new PrintWriter(outputFile);
writer.write(data.toString());
writer.close();
}
