001/* 002 * Licensed to the Apache Software Foundation (ASF) under one or more 003 * contributor license agreements. See the NOTICE file distributed with 004 * this work for additional information regarding copyright ownership. 005 * The ASF licenses this file to You under the Apache License, Version 2.0 006 * (the "License"); you may not use this file except in compliance with 007 * the License. You may obtain a copy of the License at 008 * 009 * http://www.apache.org/licenses/LICENSE-2.0 010 * 011 * Unless required by applicable law or agreed to in writing, software 012 * distributed under the License is distributed on an "AS IS" BASIS, 013 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 014 * See the License for the specific language governing permissions and 015 * limitations under the License. 016 */ 017 018package org.apache.commons.math4.legacy.analysis.function; 019 020import java.util.Arrays; 021 022import org.apache.commons.math4.legacy.analysis.ParametricUnivariateFunction; 023import org.apache.commons.math4.legacy.analysis.differentiation.DerivativeStructure; 024import org.apache.commons.math4.legacy.analysis.differentiation.UnivariateDifferentiableFunction; 025import org.apache.commons.math4.legacy.exception.DimensionMismatchException; 026import org.apache.commons.math4.legacy.exception.NotStrictlyPositiveException; 027import org.apache.commons.math4.legacy.exception.NullArgumentException; 028import org.apache.commons.math4.core.jdkmath.JdkMath; 029import org.apache.commons.numbers.core.Precision; 030 031/** 032 * <a href="http://en.wikipedia.org/wiki/Gaussian_function"> 033 * Gaussian</a> function. 034 * 035 * @since 3.0 036 */ 037public class Gaussian implements UnivariateDifferentiableFunction { 038 /** Mean. */ 039 private final double mean; 040 /** Inverse of the standard deviation. */ 041 private final double is; 042 /** Inverse of twice the square of the standard deviation. */ 043 private final double i2s2; 044 /** Normalization factor. */ 045 private final double norm; 046 047 /** 048 * Gaussian with given normalization factor, mean and standard deviation. 049 * 050 * @param norm Normalization factor. 051 * @param mean Mean. 052 * @param sigma Standard deviation. 053 * @throws NotStrictlyPositiveException if {@code sigma <= 0}. 054 */ 055 public Gaussian(double norm, 056 double mean, 057 double sigma) 058 throws NotStrictlyPositiveException { 059 if (sigma <= 0) { 060 throw new NotStrictlyPositiveException(sigma); 061 } 062 063 this.norm = norm; 064 this.mean = mean; 065 this.is = 1 / sigma; 066 this.i2s2 = 0.5 * is * is; 067 } 068 069 /** 070 * Normalized gaussian with given mean and standard deviation. 071 * 072 * @param mean Mean. 073 * @param sigma Standard deviation. 074 * @throws NotStrictlyPositiveException if {@code sigma <= 0}. 075 */ 076 public Gaussian(double mean, 077 double sigma) 078 throws NotStrictlyPositiveException { 079 this(1 / (sigma * JdkMath.sqrt(2 * Math.PI)), mean, sigma); 080 } 081 082 /** 083 * Normalized gaussian with zero mean and unit standard deviation. 084 */ 085 public Gaussian() { 086 this(0, 1); 087 } 088 089 /** {@inheritDoc} */ 090 @Override 091 public double value(double x) { 092 return value(x - mean, norm, i2s2); 093 } 094 095 /** 096 * Parametric function where the input array contains the parameters of 097 * the Gaussian. Ordered as follows: 098 * <ul> 099 * <li>Norm</li> 100 * <li>Mean</li> 101 * <li>Standard deviation</li> 102 * </ul> 103 */ 104 public static class Parametric implements ParametricUnivariateFunction { 105 /** 106 * Computes the value of the Gaussian at {@code x}. 107 * 108 * @param x Value for which the function must be computed. 109 * @param param Values of norm, mean and standard deviation. 110 * @return the value of the function. 111 * @throws NullArgumentException if {@code param} is {@code null}. 112 * @throws DimensionMismatchException if the size of {@code param} is 113 * not 3. 114 * @throws NotStrictlyPositiveException if {@code param[2]} is negative. 115 */ 116 @Override 117 public double value(double x, double ... param) 118 throws NullArgumentException, 119 DimensionMismatchException, 120 NotStrictlyPositiveException { 121 validateParameters(param); 122 123 final double diff = x - param[1]; 124 final double i2s2 = 1 / (2 * param[2] * param[2]); 125 return Gaussian.value(diff, param[0], i2s2); 126 } 127 128 /** 129 * Computes the value of the gradient at {@code x}. 130 * The components of the gradient vector are the partial 131 * derivatives of the function with respect to each of the 132 * <em>parameters</em> (norm, mean and standard deviation). 133 * 134 * @param x Value at which the gradient must be computed. 135 * @param param Values of norm, mean and standard deviation. 136 * @return the gradient vector at {@code x}. 137 * @throws NullArgumentException if {@code param} is {@code null}. 138 * @throws DimensionMismatchException if the size of {@code param} is 139 * not 3. 140 * @throws NotStrictlyPositiveException if {@code param[2]} is negative. 141 */ 142 @Override 143 public double[] gradient(double x, double ... param) 144 throws NullArgumentException, 145 DimensionMismatchException, 146 NotStrictlyPositiveException { 147 validateParameters(param); 148 149 final double norm = param[0]; 150 final double diff = x - param[1]; 151 final double sigma = param[2]; 152 final double i2s2 = 1 / (2 * sigma * sigma); 153 154 final double n = Gaussian.value(diff, 1, i2s2); 155 final double m = norm * n * 2 * i2s2 * diff; 156 final double s = m * diff / sigma; 157 158 return new double[] { n, m, s }; 159 } 160 161 /** 162 * Validates parameters to ensure they are appropriate for the evaluation of 163 * the {@link #value(double,double[])} and {@link #gradient(double,double[])} 164 * methods. 165 * 166 * @param param Values of norm, mean and standard deviation. 167 * @throws NullArgumentException if {@code param} is {@code null}. 168 * @throws DimensionMismatchException if the size of {@code param} is 169 * not 3. 170 * @throws NotStrictlyPositiveException if {@code param[2]} is negative. 171 */ 172 private void validateParameters(double[] param) 173 throws NullArgumentException, 174 DimensionMismatchException, 175 NotStrictlyPositiveException { 176 if (param == null) { 177 throw new NullArgumentException(); 178 } 179 if (param.length != 3) { 180 throw new DimensionMismatchException(param.length, 3); 181 } 182 if (param[2] <= 0) { 183 throw new NotStrictlyPositiveException(param[2]); 184 } 185 } 186 } 187 188 /** 189 * @param xMinusMean {@code x - mean}. 190 * @param norm Normalization factor. 191 * @param i2s2 Inverse of twice the square of the standard deviation. 192 * @return the value of the Gaussian at {@code x}. 193 */ 194 private static double value(double xMinusMean, 195 double norm, 196 double i2s2) { 197 return norm * JdkMath.exp(-xMinusMean * xMinusMean * i2s2); 198 } 199 200 /** {@inheritDoc} 201 * @since 3.1 202 */ 203 @Override 204 public DerivativeStructure value(final DerivativeStructure t) 205 throws DimensionMismatchException { 206 207 final double u = is * (t.getValue() - mean); 208 double[] f = new double[t.getOrder() + 1]; 209 210 // the nth order derivative of the Gaussian has the form: 211 // dn(g(x)/dxn = (norm / s^n) P_n(u) exp(-u^2/2) with u=(x-m)/s 212 // where P_n(u) is a degree n polynomial with same parity as n 213 // P_0(u) = 1, P_1(u) = -u, P_2(u) = u^2 - 1, P_3(u) = -u^3 + 3 u... 214 // the general recurrence relation for P_n is: 215 // P_n(u) = P_(n-1)'(u) - u P_(n-1)(u) 216 // as per polynomial parity, we can store coefficients of both P_(n-1) and P_n in the same array 217 final double[] p = new double[f.length]; 218 p[0] = 1; 219 final double u2 = u * u; 220 double coeff = norm * JdkMath.exp(-0.5 * u2); 221 if (coeff <= Precision.SAFE_MIN) { 222 Arrays.fill(f, 0.0); 223 } else { 224 f[0] = coeff; 225 for (int n = 1; n < f.length; ++n) { 226 227 // update and evaluate polynomial P_n(x) 228 double v = 0; 229 p[n] = -p[n - 1]; 230 for (int k = n; k >= 0; k -= 2) { 231 v = v * u2 + p[k]; 232 if (k > 2) { 233 p[k - 2] = (k - 1) * p[k - 1] - p[k - 3]; 234 } else if (k == 2) { 235 p[0] = p[1]; 236 } 237 } 238 if ((n & 0x1) == 1) { 239 v *= u; 240 } 241 242 coeff *= is; 243 f[n] = coeff * v; 244 } 245 } 246 247 return t.compose(f); 248 } 249}