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.NullArgumentException;
027import org.apache.commons.math4.core.jdkmath.JdkMath;
028
029/**
030 * <a href="http://en.wikipedia.org/wiki/Sigmoid_function">
031 *  Sigmoid</a> function.
032 * It is the inverse of the {@link Logit logit} function.
033 * A more flexible version, the generalised logistic, is implemented
034 * by the {@link Logistic} class.
035 *
036 * @since 3.0
037 */
038public class Sigmoid implements UnivariateDifferentiableFunction {
039    /** Lower asymptote. */
040    private final double lo;
041    /** Higher asymptote. */
042    private final double hi;
043
044    /**
045     * Usual sigmoid function, where the lower asymptote is 0 and the higher
046     * asymptote is 1.
047     */
048    public Sigmoid() {
049        this(0, 1);
050    }
051
052    /**
053     * Sigmoid function.
054     *
055     * @param lo Lower asymptote.
056     * @param hi Higher asymptote.
057     */
058    public Sigmoid(double lo,
059                   double hi) {
060        this.lo = lo;
061        this.hi = hi;
062    }
063
064    /** {@inheritDoc} */
065    @Override
066    public double value(double x) {
067        return value(x, lo, hi);
068    }
069
070    /**
071     * Parametric function where the input array contains the parameters of
072     * the {@link Sigmoid#Sigmoid(double,double) sigmoid function}. Ordered
073     * as follows:
074     * <ul>
075     *  <li>Lower asymptote</li>
076     *  <li>Higher asymptote</li>
077     * </ul>
078     */
079    public static class Parametric implements ParametricUnivariateFunction {
080        /**
081         * Computes the value of the sigmoid at {@code x}.
082         *
083         * @param x Value for which the function must be computed.
084         * @param param Values of lower asymptote and higher asymptote.
085         * @return the value of the function.
086         * @throws NullArgumentException if {@code param} is {@code null}.
087         * @throws DimensionMismatchException if the size of {@code param} is
088         * not 2.
089         */
090        @Override
091        public double value(double x, double ... param)
092            throws NullArgumentException,
093                   DimensionMismatchException {
094            validateParameters(param);
095            return Sigmoid.value(x, param[0], param[1]);
096        }
097
098        /**
099         * Computes the value of the gradient at {@code x}.
100         * The components of the gradient vector are the partial
101         * derivatives of the function with respect to each of the
102         * <em>parameters</em> (lower asymptote and higher asymptote).
103         *
104         * @param x Value at which the gradient must be computed.
105         * @param param Values for lower asymptote and higher asymptote.
106         * @return the gradient vector at {@code x}.
107         * @throws NullArgumentException if {@code param} is {@code null}.
108         * @throws DimensionMismatchException if the size of {@code param} is
109         * not 2.
110         */
111        @Override
112        public double[] gradient(double x, double ... param)
113            throws NullArgumentException,
114                   DimensionMismatchException {
115            validateParameters(param);
116
117            final double invExp1 = 1 / (1 + JdkMath.exp(-x));
118
119            return new double[] { 1 - invExp1, invExp1 };
120        }
121
122        /**
123         * Validates parameters to ensure they are appropriate for the evaluation of
124         * the {@link #value(double,double[])} and {@link #gradient(double,double[])}
125         * methods.
126         *
127         * @param param Values for lower and higher asymptotes.
128         * @throws NullArgumentException if {@code param} is {@code null}.
129         * @throws DimensionMismatchException if the size of {@code param} is
130         * not 2.
131         */
132        private void validateParameters(double[] param)
133            throws NullArgumentException,
134                   DimensionMismatchException {
135            if (param == null) {
136                throw new NullArgumentException();
137            }
138            if (param.length != 2) {
139                throw new DimensionMismatchException(param.length, 2);
140            }
141        }
142    }
143
144    /**
145     * @param x Value at which to compute the sigmoid.
146     * @param lo Lower asymptote.
147     * @param hi Higher asymptote.
148     * @return the value of the sigmoid function at {@code x}.
149     */
150    private static double value(double x,
151                                double lo,
152                                double hi) {
153        return lo + (hi - lo) / (1 + JdkMath.exp(-x));
154    }
155
156    /** {@inheritDoc}
157     * @since 3.1
158     */
159    @Override
160    public DerivativeStructure value(final DerivativeStructure t)
161        throws DimensionMismatchException {
162
163        double[] f = new double[t.getOrder() + 1];
164        final double exp = JdkMath.exp(-t.getValue());
165        if (Double.isInfinite(exp)) {
166
167            // special handling near lower boundary, to avoid NaN
168            f[0] = lo;
169            Arrays.fill(f, 1, f.length, 0.0);
170        } else {
171
172            // the nth order derivative of sigmoid has the form:
173            // dn(sigmoid(x)/dxn = P_n(exp(-x)) / (1+exp(-x))^(n+1)
174            // where P_n(t) is a degree n polynomial with normalized higher term
175            // P_0(t) = 1, P_1(t) = t, P_2(t) = t^2 - t, P_3(t) = t^3 - 4 t^2 + t...
176            // the general recurrence relation for P_n is:
177            // P_n(x) = n t P_(n-1)(t) - t (1 + t) P_(n-1)'(t)
178            final double[] p = new double[f.length];
179
180            final double inv   = 1 / (1 + exp);
181            double coeff = hi - lo;
182            for (int n = 0; n < f.length; ++n) {
183
184                // update and evaluate polynomial P_n(t)
185                double v = 0;
186                p[n] = 1;
187                for (int k = n; k >= 0; --k) {
188                    v = v * exp + p[k];
189                    if (k > 1) {
190                        p[k - 1] = (n - k + 2) * p[k - 2] - (k - 1) * p[k - 1];
191                    } else {
192                        p[0] = 0;
193                    }
194                }
195
196                coeff *= inv;
197                f[n]   = coeff * v;
198            }
199
200            // fix function value
201            f[0] += lo;
202        }
203
204        return t.compose(f);
205    }
206}