MultiStartUnivariateOptimizer.java

/*
 * Licensed to the Apache Software Foundation (ASF) under one or more
 * contributor license agreements.  See the NOTICE file distributed with
 * this work for additional information regarding copyright ownership.
 * The ASF licenses this file to You 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
 *
 *      https://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.
 */

/*
 * This is not the original file distributed by the Apache Software Foundation
 * It has been modified by the Hipparchus project
 */

package org.hipparchus.optim.univariate;

import java.util.Arrays;
import java.util.Comparator;

import org.hipparchus.exception.LocalizedCoreFormats;
import org.hipparchus.exception.MathIllegalStateException;
import org.hipparchus.exception.MathIllegalArgumentException;
import org.hipparchus.optim.MaxEval;
import org.hipparchus.optim.OptimizationData;
import org.hipparchus.optim.nonlinear.scalar.GoalType;
import org.hipparchus.random.RandomGenerator;

/**
 * Special implementation of the {@link UnivariateOptimizer} interface
 * adding multi-start features to an existing optimizer.
 * <br>
 * This class wraps an optimizer in order to use it several times in
 * turn with different starting points (trying to avoid being trapped
 * in a local extremum when looking for a global one).
 *
 */
public class MultiStartUnivariateOptimizer
    extends UnivariateOptimizer {
    /** Underlying classical optimizer. */
    private final UnivariateOptimizer optimizer;
    /** Number of evaluations already performed for all starts. */
    private int totalEvaluations;
    /** Number of starts to go. */
    private final int starts;
    /** Random generator for multi-start. */
    private final RandomGenerator generator;
    /** Found optima. */
    private UnivariatePointValuePair[] optima;
    /** Optimization data. */
    private OptimizationData[] optimData;
    /**
     * Location in {@link #optimData} where the updated maximum
     * number of evaluations will be stored.
     */
    private int maxEvalIndex = -1;
    /**
     * Location in {@link #optimData} where the updated start value
     * will be stored.
     */
    private int searchIntervalIndex = -1;

    /**
     * Create a multi-start optimizer from a single-start optimizer.
     *
     * @param optimizer Single-start optimizer to wrap.
     * @param starts Number of starts to perform. If {@code starts == 1},
     * the {@code optimize} methods will return the same solution as
     * {@code optimizer} would.
     * @param generator Random generator to use for restarts.
     * @throws MathIllegalArgumentException if {@code starts < 1}.
     */
    public MultiStartUnivariateOptimizer(final UnivariateOptimizer optimizer,
                                         final int starts,
                                         final RandomGenerator generator) {
        super(optimizer.getConvergenceChecker());

        if (starts < 1) {
            throw new MathIllegalArgumentException(LocalizedCoreFormats.NUMBER_TOO_SMALL,
                                                   starts, 1);
        }

        this.optimizer = optimizer;
        this.starts = starts;
        this.generator = generator;
    }

    /** {@inheritDoc} */
    @Override
    public int getEvaluations() {
        return totalEvaluations;
    }

    /**
     * Gets all the optima found during the last call to {@code optimize}.
     * The optimizer stores all the optima found during a set of
     * restarts. The {@code optimize} method returns the best point only.
     * This method returns all the points found at the end of each starts,
     * including the best one already returned by the {@code optimize} method.
     * <br>
     * The returned array as one element for each start as specified
     * in the constructor. It is ordered with the results from the
     * runs that did converge first, sorted from best to worst
     * objective value (i.e in ascending order if minimizing and in
     * descending order if maximizing), followed by {@code null} elements
     * corresponding to the runs that did not converge. This means all
     * elements will be {@code null} if the {@code optimize} method did throw
     * an exception.
     * This also means that if the first element is not {@code null}, it is
     * the best point found across all starts.
     *
     * @return an array containing the optima.
     * @throws MathIllegalStateException if {@link #optimize(OptimizationData[])
     * optimize} has not been called.
     */
    public UnivariatePointValuePair[] getOptima() {
        if (optima == null) {
            throw new MathIllegalStateException(LocalizedCoreFormats.NO_OPTIMUM_COMPUTED_YET);
        }
        return optima.clone();
    }

    /**
     * {@inheritDoc}
     *
     * @throws MathIllegalStateException if {@code optData} does not contain an
     * instance of {@link MaxEval} or {@link SearchInterval}.
     */
    @Override
    public UnivariatePointValuePair optimize(OptimizationData... optData) {
        // Store arguments in order to pass them to the internal optimizer.
       optimData = optData.clone();
        // Set up base class and perform computations.
        return super.optimize(optData);
    }

    /** {@inheritDoc} */
    @Override
    protected UnivariatePointValuePair doOptimize() {
        // Remove all instances of "MaxEval" and "SearchInterval" from the
        // array that will be passed to the internal optimizer.
        // The former is to enforce smaller numbers of allowed evaluations
        // (according to how many have been used up already), and the latter
        // to impose a different start value for each start.
        for (int i = 0; i < optimData.length; i++) {
            if (optimData[i] instanceof MaxEval) {
                optimData[i] = null;
                maxEvalIndex = i;
                continue;
            }
            if (optimData[i] instanceof SearchInterval) {
                optimData[i] = null;
                searchIntervalIndex = i;
                continue;
            }
        }
        if (maxEvalIndex == -1) {
            throw new MathIllegalStateException(LocalizedCoreFormats.ILLEGAL_STATE);
        }
        if (searchIntervalIndex == -1) {
            throw new MathIllegalStateException(LocalizedCoreFormats.ILLEGAL_STATE);
        }

        RuntimeException lastException = null;
        optima = new UnivariatePointValuePair[starts];
        totalEvaluations = 0;

        final int maxEval = getMaxEvaluations();
        final double min = getMin();
        final double max = getMax();
        final double startValue = getStartValue();

        // Multi-start loop.
        for (int i = 0; i < starts; i++) {
            // CHECKSTYLE: stop IllegalCatch
            try {
                // Decrease number of allowed evaluations.
                optimData[maxEvalIndex] = new MaxEval(maxEval - totalEvaluations);
                // New start value.
                final double s = (i == 0) ?
                    startValue :
                    min + generator.nextDouble() * (max - min);
                optimData[searchIntervalIndex] = new SearchInterval(min, max, s);
                // Optimize.
                optima[i] = optimizer.optimize(optimData);
            } catch (RuntimeException mue) { // NOPMD - caching a RuntimeException is intentional here, it will be rethrown later
                lastException = mue;
                optima[i] = null;
            }
            // CHECKSTYLE: resume IllegalCatch

            totalEvaluations += optimizer.getEvaluations();
        }

        sortPairs(getGoalType());

        if (optima[0] == null) {
            throw lastException; // Cannot be null if starts >= 1.
        }

        // Return the point with the best objective function value.
        return optima[0];
    }

    /**
     * Sort the optima from best to worst, followed by {@code null} elements.
     *
     * @param goal Goal type.
     */
    private void sortPairs(final GoalType goal) {
        Arrays.sort(optima, new Comparator<UnivariatePointValuePair>() {
                /** {@inheritDoc} */
                @Override
                public int compare(final UnivariatePointValuePair o1,
                                   final UnivariatePointValuePair o2) {
                    if (o1 == null) {
                        return (o2 == null) ? 0 : 1;
                    } else if (o2 == null) {
                        return -1;
                    }
                    final double v1 = o1.getValue();
                    final double v2 = o2.getValue();
                    return (goal == GoalType.MINIMIZE) ?
                        Double.compare(v1, v2) : Double.compare(v2, v1);
                }
            });
    }
}