mikera.matrixx.decompose.impl.qr.HouseholderQR Maven / Gradle / Ivy
Show all versions of vectorz Show documentation
/*
* Copyright (c) 2009-2013, Peter Abeles. All Rights Reserved.
*
* This file is part of Efficient Java Matrix Library (EJML).
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package mikera.matrixx.decompose.impl.qr;
import mikera.matrixx.AMatrix;
import mikera.matrixx.Matrix;
/**
*
* This variation of QR decomposition uses reflections to compute the Q matrix.
* Each reflection uses a householder operations, hence its name. To provide a meaningful solution
* the original matrix must have full rank. This is intended for processing of small to medium matrices.
*
*
* Both Q and R are stored in the same m by n matrix. Q is not stored directly, instead the u from
* Qk=(I-γ*u*uT) is stored. Decomposition requires about 2n*m2-2m2/3 flops.
*
*
*
* See the QR reflections algorithm described in:
* David S. Watkins, "Fundamentals of Matrix Computations" 2nd Edition, 2002
*
*
*
* For the most part this is a straight forward implementation. To improve performance on large matrices a column is written to an array and the order
* of some of the loops has been changed. This will degrade performance noticeably on small matrices. Since
* it is unlikely that the QR decomposition would be a bottle neck when small matrices are involved only
* one implementation is provided.
*
*
* @author Peter Abeles
*/
public class HouseholderQR implements QRDecomposition {
/**
* Where the Q and R matrices are stored. R is stored in the
* upper triangular portion and Q on the lower bit. Lower columns
* are where u is stored. Q_k = (I - gamma_k*u_k*u_k^T).
*/
protected Matrix QR;
// used internally to store temporary data
protected double u[],v[];
// dimension of the decomposed matrices
protected int numCols; // this is 'n'
protected int numRows; // this is 'm'
protected int minLength;
protected double dataQR[];
// the computed gamma for Q_k matrix
protected double gammas[];
// local variables
protected double gamma;
protected double tau;
// did it encounter an error?
protected boolean error;
private boolean compact;
private AMatrix Q;
private AMatrix R;
public HouseholderQR(boolean compact) {
this.compact = compact;
}
/**
* Returns a single matrix which contains the combined values of Q and R. This
* is possible since Q is symmetric and R is upper triangular.
*
* @return The combined Q R matrix.
*/
public AMatrix getQR() {
return QR;
}
/**
* @return The Q matrix from the decomposition.
*/
public AMatrix getQ() {
if (Q == null) {
Q = computeQ();
}
return Q;
}
/**
* @return The R matrix from the decomposition.
*/
public AMatrix getR() {
if (R == null) {
R = computeR();
}
return R;
}
/**
* Computes the Q matrix from the information stored in the QR matrix. This
* operation requires about 4(m2n-mn2+n3/3) flops.
*/
protected AMatrix computeQ() {
Matrix Q = Matrix.createIdentity(numRows);
for( int j = minLength-1; j >= 0; j-- ) {
u[j] = 1;
for( int i = j+1; i < numRows; i++ ) {
u[i] = QR.get(i,j);
}
QRHelperFunctions.rank1UpdateMultR(Q,u,gammas[j],j,j,numRows,v);
}
return Q;
}
/**
* Returns an upper triangular matrix which is the R in the QR decomposition.
*/
protected AMatrix computeR() {
Matrix R;
if( compact ) {
R = Matrix.create(minLength,numCols);
} else {
R = Matrix.create(numRows,numCols);
}
for( int i = 0; i < minLength; i++ ) {
for( int j = i; j < numCols; j++ ) {
double val = QR.get(i,j);
R.set(i,j,val);
}
}
return R;
}
/**
*
* In order to decompose the matrix 'A' it must have full rank. 'A' is a 'm' by 'n' matrix.
* It requires about 2n*m2-2m2/3 flops.
*
*
*
* The matrix provided here can be of different
* dimension than the one specified in the constructor. It just has to be smaller than or equal
* to it.
*
*/
@Override
public QRResult decompose( AMatrix A ) {
error = false;
this.numCols = A.columnCount();
this.numRows = A.rowCount();
minLength = Math.min(numRows,numCols);
int maxLength = Math.max(numRows,numCols);
QR = Matrix.create(A);
u = new double[ maxLength ];
v = new double[ maxLength ];
dataQR = QR.data;
gammas = new double[ minLength ];
for( int j = 0; j < minLength; j++ ) {
householder(j);
updateA(j);
}
// if (error) return null; // TODO: figure out how to handle
return new QRResult(getQ(), getR());
}
/**
*
* Computes the householder vector "u" for the first column of submatrix j. Note this is
* a specialised householder for this problem. There is some protection against
* overflow and underflow.
*
*
* Q = I - γuuT
*
*
* This function finds the values of 'u' and 'γ'.
*
*
* @param j Which submatrix to work off of.
*/
protected void householder( int j )
{
// find the element with the largest absolute value in the column and make a copy
int index = j+j*numCols;
double max = 0;
for( int i = j; i < numRows; i++ ) {
double d = u[i] = dataQR[index];
// absolute value of d
if( d < 0 ) d = -d;
if( max < d ) {
max = d;
}
index += numCols;
}
if( max == 0.0 ) {
gamma = 0;
error = true;
} else {
// compute the norm2 of the matrix, with each element
// normalized by the max value to avoid overflow problems
tau = 0;
for( int i = j; i < numRows; i++ ) {
u[i] /= max;
double d = u[i];
tau += d*d;
}
tau = Math.sqrt(tau);
if( u[j] < 0 )
tau = -tau;
double u_0 = u[j] + tau;
gamma = u_0/tau;
for( int i = j+1; i < numRows; i++ ) {
u[i] /= u_0;
}
u[j] = 1;
tau *= max;
}
gammas[j] = gamma;
}
/**
*
* Takes the results from the householder computation and updates the 'A' matrix.
*
* A = (I - γ*u*uT)A
*
*
* @param w The submatrix.
*/
protected void updateA( int w )
{
// much of the code below is equivalent to the rank1Update function
// however, since τ has already been computed there is no need to
// recompute it, saving a few multiplication operations
// for( int i = w+1; i < numCols; i++ ) {
// double val = 0;
//
// for( int k = w; k < numRows; k++ ) {
// val += u[k]*dataQR[k*numCols +i];
// }
// v[i] = gamma*val;
// }
// This is functionally the same as the above code but the order has been changed
// to avoid jumping the cpu cache
for( int i = w+1; i < numCols; i++ ) {
v[i] = u[w]*dataQR[w*numCols +i];
}
for( int k = w+1; k < numRows; k++ ) {
int indexQR = k*numCols+w+1;
for( int i = w+1; i < numCols; i++ ) {
// v[i] += u[k]*dataQR[k*numCols +i];
v[i] += u[k]*dataQR[indexQR++];
}
}
for( int i = w+1; i < numCols; i++ ) {
v[i] *= gamma;
}
// end of reordered code
for( int i = w; i < numRows; i++ ) {
double valU = u[i];
int indexQR = i*numCols+w+1;
for( int j = w+1; j < numCols; j++ ) {
// dataQR[i*numCols+j] -= valU*v[j];
dataQR[indexQR++] -= valU*v[j];
}
}
if( w < numCols ) {
dataQR[w+w*numCols] = -tau;
}
// save the Q matrix in the lower portion of QR
for( int i = w+1; i < numRows; i++ ) {
dataQR[w+i*numCols] = u[i];
}
}
public double[] getGammas() {
return gammas;
}
}