/**
* Computes the d and H parameters. Where d is the average error gradient and
* H is an approximation of the hessian.
*/
private void computeDandH(DMatrixRMaj param, DMatrixRMaj x, DMatrixRMaj y)
{
func.compute(param, x, tempDH);
CommonOps_DDRM.subtractEquals(tempDH, y);
computeNumericalJacobian(param, x, jacobian);
int numParam = param.getNumElements();
int length = x.getNumElements();
// d = average{ (f(x_i;p) - y_i) * jacobian(:,i) }
for (int i = 0; i < numParam; i++)
{
double total = 0;
for (int j = 0; j < length; j++)
{
total += tempDH.get(j, 0) * jacobian.get(i, j);
}
d.set(i, 0, total / length);
}
// compute the approximation of the hessian
CommonOps_DDRM.multTransB(jacobian, jacobian, H);
CommonOps_DDRM.scale(1.0 / length, H);
}
//@Override
public void update(DMatrixRMaj z, DMatrixRMaj R)
{
// y = z - H x
CommonOps_DDRM.mult(H, x, y);
CommonOps_DDRM.subtract(z, y, y);
// S = H P H' + R
CommonOps_DDRM.mult(H, P, c);
CommonOps_DDRM.multTransB(c, H, S);
CommonOps_DDRM.addEquals(S, R);
// K = PH'S^(-1)
if (!solver.setA(S))
{
throw new InvalidOperationException("Invert failed");
}
solver.invert(S_inv);
CommonOps_DDRM.multTransA(H, S_inv, d);
CommonOps_DDRM.mult(P, d, K);
// x = x + Ky
CommonOps_DDRM.mult(K, y, a);
CommonOps_DDRM.addEquals(x, a);
// P = (I-kH)P = P - (KH)P = P-K(HP)
CommonOps_DDRM.mult(H, P, c);
CommonOps_DDRM.mult(K, c, b);
CommonOps_DDRM.subtractEquals(P, b);
}
//@Override
public void predict()
{
// x = F x
CommonOps_DDRM.mult(F, x, a);
x.set(a);
// P = F P F' + Q
CommonOps_DDRM.mult(F, P, b);
CommonOps_DDRM.multTransB(b, F, P);
CommonOps_DDRM.addEquals(P, Q);
}
/**
* Creates a newJava.Util.Random symmetric matrix that will have the specified real eigenvalues.
*
* @param num Dimension of the resulting matrix.
* @param randJava.Util.Random number generator.
* @param eigenvalues Set of real eigenvalues that the matrix will have.
* @return AJava.Util.Random matrix with the specified eigenvalues.
*/
public static DMatrixRMaj symmetricWithEigenvalues(int num, Java.Util.Random rand, double[] eigenvalues)
{
DMatrixRMaj V = RandomMatrices_DDRM.orthogonal(num, num, rand);
DMatrixRMaj D = CommonOps_DDRM.diag(eigenvalues);
DMatrixRMaj temp = new DMatrixRMaj(num, num);
CommonOps_DDRM.mult(V, D, temp);
CommonOps_DDRM.multTransB(temp, V, D);
return(D);
}
/**
* Creates aJava.Util.Random symmetric positive definite matrix.
*
* @param width The width of the square matrix it returns.
* @param randJava.Util.Random number generator used to make the matrix.
* @return TheJava.Util.Random symmetric positive definite matrix.
*/
public static DMatrixRMaj symmetricPosDef(int width, Java.Util.Random rand)
{
// This is not formally proven to work. It just seems to work.
DMatrixRMaj a = new DMatrixRMaj(width, 1);
DMatrixRMaj b = new DMatrixRMaj(width, width);
for (int i = 0; i < width; i++)
{
a.set(i, 0, rand.NextDouble());
}
CommonOps_DDRM.multTransB(a, a, b);
for (int i = 0; i < width; i++)
{
b.add(i, i, 1);
}
return(b);
}
/**
* <p>
* Creates aJava.Util.Random matrix which will have the provided singular values. The Count() of sv
* is assumed to be the rank of the matrix. This can be useful for testing purposes when one
* needs to ensure that a matrix is not singular but randomly generated.
* </p>
*
* @param numRows Number of rows in generated matrix.
* @param numCols NUmber of columns in generated matrix.
* @param randJava.Util.Random number generator.
* @param sv Singular values of the matrix.
* @return A new matrix with the specified singular values.
*/
public static DMatrixRMaj singular(int numRows, int numCols,
Java.Util.Random rand, double[] sv)
{
DMatrixRMaj U, V, S;
// speed it up in compact format
if (numRows > numCols)
{
U = RandomMatrices_DDRM.orthogonal(numRows, numCols, rand);
V = RandomMatrices_DDRM.orthogonal(numCols, numCols, rand);
S = new DMatrixRMaj(numCols, numCols);
}
else
{
U = RandomMatrices_DDRM.orthogonal(numRows, numRows, rand);
V = RandomMatrices_DDRM.orthogonal(numCols, numCols, rand);
S = new DMatrixRMaj(numRows, numCols);
}
int min = Math.Min(numRows, numCols);
min = Math.Min(min, sv.Count());
for (int i = 0; i < min; i++)
{
S.set(i, i, sv[i]);
}
DMatrixRMaj tmp = new DMatrixRMaj(numRows, numCols);
CommonOps_DDRM.mult(U, S, tmp);
S.reshape(numRows, numCols);
CommonOps_DDRM.multTransB(tmp, V, S);
return(S);
}
/**
* <p>
* Creates a random matrix which will have the provided singular values. The length of sv
* is assumed to be the rank of the matrix. This can be useful for testing purposes when one
* needs to ensure that a matrix is not singular but randomly generated.
* </p>
*
* @param numRows Number of rows in generated matrix.
* @param numCols NUmber of columns in generated matrix.
* @param rand Random number generator.
* @param sv Singular values of the matrix.
* @return A new matrix with the specified singular values.
*/
public static DMatrixRMaj singleValues(int numRows, int numCols,
IMersenneTwister rand, double[] sv)
{
DMatrixRMaj U = RandomMatrices_DDRM.orthogonal(numRows, numRows, rand);
DMatrixRMaj V = RandomMatrices_DDRM.orthogonal(numCols, numCols, rand);
DMatrixRMaj S = new DMatrixRMaj(numRows, numCols);
int min = Math.Min(numRows, numCols);
min = Math.Min(min, sv.Length);
for (int i = 0; i < min; i++)
{
S.set(i, i, sv[i]);
}
DMatrixRMaj tmp = new DMatrixRMaj(numRows, numCols);
CommonOps_DDRM.mult(U, S, tmp);
CommonOps_DDRM.multTransB(tmp, V, S);
return(S);
}
public virtual bool decompose(DMatrixRMaj orig)
{
if (!decompQRP.decompose(orig))
{
return(false);
}
m = orig.numRows;
n = orig.numCols;
min = Math.Min(m, n);
B.reshape(min, n, false);
decompQRP.getR(B, true);
// apply the column pivots.
// TODO this is horribly inefficient
DMatrixRMaj result = new DMatrixRMaj(min, n);
DMatrixRMaj P = decompQRP.getColPivotMatrix(null);
CommonOps_DDRM.multTransB(B, P, result);
B.set(result);
return(decompBi.decompose(B));
}