org.cicirello.search.problems.Plateaus Maven / Gradle / Ivy

Go to download

Show more of this group Show more artifacts with this name
Show all versions of chips-n-salsa Show documentation

Chips-n-Salsa is a Java library of customizable, hybridizable, iterative, parallel, stochastic, and self-adaptive local search algorithms. The library includes implementations of several stochastic local search algorithms, including simulated annealing, hill climbers, as well as constructive search algorithms such as stochastic sampling. Chips-n-Salsa now also includes genetic algorithms as well as evolutionary algorithms more generally. The library very extensively supports simulated annealing. It includes several classes for representing solutions to a variety of optimization problems. For example, the library includes a BitVector class that implements vectors of bits, as well as classes for representing solutions to problems where we are searching for an optimal vector of integers or reals. For each of the built-in representations, the library provides the most common mutation operators for generating random neighbors of candidate solutions, as well as common crossover operators for use with evolutionary algorithms. Additionally, the library provides extensive support for permutation optimization problems, including implementations of many different mutation operators for permutations, and utilizing the efficiently implemented Permutation class of the JavaPermutationTools (JPT) library. Chips-n-Salsa is customizable, making extensive use of Java's generic types, enabling using the library to optimize other types of representations beyond what is provided in the library. It is hybridizable, providing support for integrating multiple forms of local search (e.g., using a hill climber on a solution generated by simulated annealing), creating hybrid mutation operators (e.g., local search using multiple mutation operators), as well as support for running more than one type of search for the same problem concurrently using multiple threads as a form of algorithm portfolio. Chips-n-Salsa is iterative, with support for multistart metaheuristics, including implementations of several restart schedules for varying the run lengths across the restarts. It also supports parallel execution of multiple instances of the same, or different, stochastic local search algorithms for an instance of a problem to accelerate the search process. The library supports self-adaptive search in a variety of ways, such as including implementations of adaptive annealing schedules for simulated annealing, such as the Modified Lam schedule, implementations of the simpler annealing schedules but which self-tune the initial temperature and other parameters, and restart schedules that adapt to run length.

There is a newer version: 7.0.1

Show newest version

/*
 * Chips-n-Salsa: A library of parallel self-adaptive local search algorithms.
 * Copyright (C) 2002-2021  Vincent A. Cicirello
 *
 * This file is part of Chips-n-Salsa (https://chips-n-salsa.cicirello.org/).
 *
 * Chips-n-Salsa is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * Chips-n-Salsa is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program.  If not, see .
 */

package org.cicirello.search.problems;

import org.cicirello.search.representations.BitVector;

/**
 * This class implements Ackley's Plateaus problem, an artificial search landscape over the space of
 * bitstrings that is characterized by large flat regions known as plateaus. This is specifically an
 * implementation of Ackley's 1987 version of the problem (he described a similar problem in an
 * earlier 1985 paper).
 *
 * The Plateaus problem involves maximizing the following function. Divide the bits of the bit
 * string into four equal sized parts. For each of the four parts, check whether all bits in the
 * segment are equal to a 1, and if so, then that segment contributes 2.5*n to the fitness function,
 * where n is the length of the entire bit string (if there are any 0s in the segment, then that
 * segment doesn't contribute anything to the fitness function). Since there are four segments the
 * optimum occurs when the entire bit string is all 1s, which has a maximum fitness of 10*n. The
 * entire search space only has 5 possible fitness values: 0, 2.5*n, 5*n, 7.5*n, and 10*n.
 *
 * 
The {@link #value value} method implements the original maximization version of the Plateaus
 * problem, as described above. The algorithms of the Chips-n-Salsa library are defined for
 * minimization, requiring a cost function. The {@link #cost cost} method implements the equivalent
 * as the following minimization problem: minimize cost(x) = 10*n - f(x), where f(x) is the Plateaus
 * function as defined above. The global optima is still all 1-bits, which has a cost equal to 0.
 *
 * The Plateaus problem was introduced by David Ackley in the following paper:

 * David H. Ackley. An empirical study of bit vector function optimization. Genetic Algorithms and
 * Simulated Annealing, pages 170-204, 1987.
 *
 * @author Vincent A. Cicirello, https://www.cicirello.org/
 * @version 3.25.2021
 */
public final class Plateaus implements OptimizationProblem {

  /** Constructs an instance of Ackley's Plateaus problem. */
  public Plateaus() {}

  @Override
  public double cost(BitVector candidate) {
    return 10 * candidate.length() - value(candidate);
  }

  @Override
  public double minCost() {
    return 0;
  }

  @Override
  public double value(BitVector candidate) {
    // Segment size
    int m = candidate.length() >> 2;
    // Num segments with an extra bit if n not divisible by 4
    int r = candidate.length() & 3;
    int blockCount = 0;
    BitVector.BitIterator iter = candidate.bitIterator(32);
    for (int i = r; i < 4; i++) {
      if (isBlockAllOnes(iter, m)) {
        blockCount++;
      }
    }
    if (r > 0) {
      m++;
      for (int i = 0; i < r; i++) {
        if (isBlockAllOnes(iter, m)) {
          blockCount++;
        }
      }
    }
    return blockCount * candidate.length() * 2.5;
  }

  @Override
  public boolean isMinCost(double cost) {
    return cost == 0;
  }

  private boolean isBlockAllOnes(BitVector.BitIterator iter, int stillNeed) {
    while (stillNeed >= 32) {
      stillNeed -= 32;
      if (iter.nextBitBlock() != 0xffffffff) {
        iter.skip(stillNeed);
        return false;
      }
    }
    if (stillNeed > 0) {
      int mask = (1 << stillNeed) - 1;
      if (iter.nextBitBlock(stillNeed) != mask) {
        return false;
      }
    }
    return true;
  }
}