1 /* 2 * Licensed to the Apache Software Foundation (ASF) under one or more 3 * contributor license agreements. See the NOTICE file distributed with 4 * this work for additional information regarding copyright ownership. 5 * The ASF licenses this file to You under the Apache License, Version 2.0 6 * (the "License"); you may not use this file except in compliance with 7 * the License. You may obtain a copy of the License at 8 * 9 * https://www.apache.org/licenses/LICENSE-2.0 10 * 11 * Unless required by applicable law or agreed to in writing, software 12 * distributed under the License is distributed on an "AS IS" BASIS, 13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 14 * See the License for the specific language governing permissions and 15 * limitations under the License. 16 */ 17 18 /* 19 * This is not the original file distributed by the Apache Software Foundation 20 * It has been modified by the Hipparchus project 21 */ 22 23 package org.hipparchus.random; 24 25 import org.hipparchus.exception.LocalizedCoreFormats; 26 import org.hipparchus.exception.MathIllegalArgumentException; 27 import org.hipparchus.linear.RealMatrix; 28 import org.hipparchus.linear.RectangularCholeskyDecomposition; 29 30 /** 31 * A {@link RandomVectorGenerator} that generates vectors with with 32 * correlated components. 33 * <p> 34 * Random vectors with correlated components are built by combining 35 * the uncorrelated components of another random vector in such a way that 36 * the resulting correlations are the ones specified by a positive 37 * definite covariance matrix. 38 * <p> 39 * The main use for correlated random vector generation is for Monte-Carlo 40 * simulation of physical problems with several variables, for example to 41 * generate error vectors to be added to a nominal vector. A particularly 42 * interesting case is when the generated vector should be drawn from a <a 43 * href="http://en.wikipedia.org/wiki/Multivariate_normal_distribution"> 44 * Multivariate Normal Distribution</a>. The approach using a Cholesky 45 * decomposition is quite usual in this case. However, it can be extended 46 * to other cases as long as the underlying random generator provides 47 * {@link NormalizedRandomGenerator normalized values} like {@link 48 * GaussianRandomGenerator} or {@link UniformRandomGenerator}. 49 * <p> 50 * Sometimes, the covariance matrix for a given simulation is not 51 * strictly positive definite. This means that the correlations are 52 * not all independent from each other. In this case, however, the non 53 * strictly positive elements found during the Cholesky decomposition 54 * of the covariance matrix should not be negative either, they 55 * should be null. Another non-conventional extension handling this case 56 * is used here. Rather than computing <code>C = U<sup>T</sup>.U</code> 57 * where <code>C</code> is the covariance matrix and <code>U</code> 58 * is an upper-triangular matrix, we compute <code>C = B.B<sup>T</sup></code> 59 * where <code>B</code> is a rectangular matrix having 60 * more rows than columns. The number of columns of <code>B</code> is 61 * the rank of the covariance matrix, and it is the dimension of the 62 * uncorrelated random vector that is needed to compute the component 63 * of the correlated vector. This class handles this situation 64 * automatically. 65 */ 66 public class CorrelatedRandomVectorGenerator 67 implements RandomVectorGenerator { 68 /** Mean vector. */ 69 private final double[] mean; 70 /** Underlying generator. */ 71 private final NormalizedRandomGenerator generator; 72 /** Storage for the normalized vector. */ 73 private final double[] normalized; 74 /** Root of the covariance matrix. */ 75 private final RealMatrix root; 76 77 /** 78 * Builds a correlated random vector generator from its mean 79 * vector and covariance matrix. 80 * 81 * @param mean Expected mean values for all components. 82 * @param covariance Covariance matrix. 83 * @param small Diagonal elements threshold under which column are 84 * considered to be dependent on previous ones and are discarded 85 * @param generator underlying generator for uncorrelated normalized 86 * components. 87 * @throws org.hipparchus.exception.MathIllegalArgumentException 88 * if the covariance matrix is not strictly positive definite. 89 * @throws MathIllegalArgumentException if the mean and covariance 90 * arrays dimensions do not match. 91 */ 92 public CorrelatedRandomVectorGenerator(double[] mean, 93 RealMatrix covariance, double small, 94 NormalizedRandomGenerator generator) { 95 int order = covariance.getRowDimension(); 96 if (mean.length != order) { 97 throw new MathIllegalArgumentException(LocalizedCoreFormats.DIMENSIONS_MISMATCH, 98 mean.length, order); 99 } 100 this.mean = mean.clone(); 101 102 final RectangularCholeskyDecomposition decomposition = 103 new RectangularCholeskyDecomposition(covariance, small); 104 root = decomposition.getRootMatrix(); 105 106 this.generator = generator; 107 normalized = new double[decomposition.getRank()]; 108 109 } 110 111 /** 112 * Builds a null mean random correlated vector generator from its 113 * covariance matrix. 114 * 115 * @param covariance Covariance matrix. 116 * @param small Diagonal elements threshold under which column are 117 * considered to be dependent on previous ones and are discarded. 118 * @param generator Underlying generator for uncorrelated normalized 119 * components. 120 * @throws org.hipparchus.exception.MathIllegalArgumentException 121 * if the covariance matrix is not strictly positive definite. 122 */ 123 public CorrelatedRandomVectorGenerator(RealMatrix covariance, double small, 124 NormalizedRandomGenerator generator) { 125 int order = covariance.getRowDimension(); 126 mean = new double[order]; 127 for (int i = 0; i < order; ++i) { 128 mean[i] = 0; 129 } 130 131 final RectangularCholeskyDecomposition decomposition = 132 new RectangularCholeskyDecomposition(covariance, small); 133 root = decomposition.getRootMatrix(); 134 135 this.generator = generator; 136 normalized = new double[decomposition.getRank()]; 137 138 } 139 140 /** Get the underlying normalized components generator. 141 * @return underlying uncorrelated components generator 142 */ 143 public NormalizedRandomGenerator getGenerator() { 144 return generator; 145 } 146 147 /** Get the rank of the covariance matrix. 148 * The rank is the number of independent rows in the covariance 149 * matrix, it is also the number of columns of the root matrix. 150 * @return rank of the square matrix. 151 * @see #getRootMatrix() 152 */ 153 public int getRank() { 154 return normalized.length; 155 } 156 157 /** Get the root of the covariance matrix. 158 * The root is the rectangular matrix <code>B</code> such that 159 * the covariance matrix is equal to <code>B.B<sup>T</sup></code> 160 * @return root of the square matrix 161 * @see #getRank() 162 */ 163 public RealMatrix getRootMatrix() { 164 return root; 165 } 166 167 /** Generate a correlated random vector. 168 * @return a random vector as an array of double. The returned array 169 * is created at each call, the caller can do what it wants with it. 170 */ 171 @Override 172 public double[] nextVector() { 173 174 // generate uncorrelated vector 175 for (int i = 0; i < normalized.length; ++i) { 176 normalized[i] = generator.nextNormalizedDouble(); 177 } 178 179 // compute correlated vector 180 double[] correlated = new double[mean.length]; 181 for (int i = 0; i < correlated.length; ++i) { 182 correlated[i] = mean[i]; 183 for (int j = 0; j < root.getColumnDimension(); ++j) { 184 correlated[i] += root.getEntry(i, j) * normalized[j]; 185 } 186 } 187 188 return correlated; 189 } 190 191 }