org.cicirello.search.problems.PermutationToBitVectorProblem 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) 2002-2022 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.permutations.Permutation;
import org.cicirello.search.operators.Initializer;
import org.cicirello.search.operators.bits.BitVectorInitializer;
import org.cicirello.search.representations.BitVector;

/**
 * This class implements a mapping between Permutation problems and BitVector problems, enabling
 * using {@link BitVector BitVector} search operators to solve problems defined over the space of
 * {@link Permutation Permutation} objects. It can also be used as an {@link Initializer} of
 * BitVectors by search algorithms to ensure that the search is using BitVectors of the appropriate
 * length to represent permutations of the desired length for the problem you are solving. In fact,
 * to ensure that your search is using the correct bit length, you should use this as your
 * Initializer.
 *
 * The mapping uses a classic transformation from vector of bits to a permutation of the N
 * integers from 0 to N-1 that works as follows. Consider a list of integers L initially containing
 * the N integers in sorted order: L = [0, 1, 2, ..., N-1]. The list L is circular, such that any
 * integer is an index into L. For example, if L is of length 5, indexes 0, 5, 10, etc all refer to
 * the element in position 0. Likewise, indexes 1, 6, 11, etc all refer to the element in position
 * 1, and so forth. The bit vector must be length: (N - 1) ceiling(lg N). Each contiguous group of
 * ceiling(lg N) bits is treated as an index into L. There are N-1 such indexes available based on
 * the bit vector length. Each such index is used to select an element from L, add it to the next
 * available position of an initially empty permutation P, and remove it from L. After N-1 such
 * operations, there will be one element left in L, which is then added to the end. The {@link
 * #toPermutation toPermutation(BitVector)} method implements this transformation.
 *
 * 
Here is an example. Consider a very small permutation length N = 6. For this length
 * permutation, the bit vector must be (6 - 1) ceiling(lg 6) = 5 * 3 = 15 bits in length. Let's
 * consider the transformation from the bit vector 111110001010100.
 *
 * 

 *   L is initially L = [0, 1, 2, 3, 4, 5], and P is initially P = [].
 *   
The first 3 bits, 111, is 7 in decimal, and 7 mod 6 = 1. L[1]=1, so we remove the 1 from L
 *       and add it to the end of P. This leads to L = [0, 2, 3, 4, 5], and P is P = [1].
 *   
The next 3 bits, 110, is 6 in decimal, and 6 mod 5 = 1. L[1]=2, so we remove the 2 from L
 *       and add it to the end of P. This leads to L = [0, 3, 4, 5], and P is P = [1, 2].
 *   
The next 3 bits, 001, is 1 in decimal, and 1 mod 4 = 1. L[1]=3, so we remove the 3 from L
 *       and add it to the end of P. This leads to L = [0, 4, 5], and P is P = [1, 2, 3].
 *   
The next 3 bits, 010, is 2 in decimal, and 2 mod 3 = 2. L[2]=5, so we remove the 5 from L
 *       and add it to the end of P. This leads to L = [0, 4], and P is P = [1, 2, 3, 5].
 *   
The last 3 bits, 100, is 4 in decimal, and 4 mod 2 = 0. L[0]=0, so we remove the 0 from L
 *       and add it to the end of P. This leads to L = [4], and P is P = [1, 2, 3, 5, 0].
 *   
Reached the end of the bit vector, and there is only one element left in L, the 4, which we
 *       simply add to the end of P to arrive at P = [1, 2, 3, 5, 0, 4].
 * 
 *
 * This class has two nested subclasses, {@link DoubleCost} and {@link IntegerCost}, that handle
 * the transformations from Permutation optimization problems with costs of type double and int,
 * respectively (i.e., classes {@link OptimizationProblem} and {@link
 * IntegerCostOptimizationProblem}).
 *
 * @author Vincent A. Cicirello, https://www.cicirello.org/
 */
public class PermutationToBitVectorProblem implements Initializer {

  private final BitVectorInitializer init;
  private final int bitsPerElement;
  private final int permutationLength;

  /**
   * Initializes the PermutationToBitVectorProblem mapping for a specific permutation length.
   *
   * @param permutationLength The length of the permutations under optimization, in number of
   *     elements. This is NOT the length of the BitVectors. For example, if the problem is the
   *     Traveling Salesperson, and if the instance has 100 cities, then you would pass 100 for this
   *     parameter.
   * @throws IllegalArgumentException if permutationLength is less than 1.
   */
  public PermutationToBitVectorProblem(int permutationLength) {
    if (permutationLength < 1)
      throw new IllegalArgumentException("permutationLength must be positive");
    bitsPerElement = 32 - Integer.numberOfLeadingZeros(permutationLength - 1);
    init = new BitVectorInitializer(bitsPerElement * (permutationLength - 1));
    this.permutationLength = permutationLength;
  }

  /*
   * package private for use by split
   */
  PermutationToBitVectorProblem(PermutationToBitVectorProblem other) {
    init = other.init.split();
    bitsPerElement = other.bitsPerElement;
    permutationLength = other.permutationLength;
  }

  @Override
  public final BitVector createCandidateSolution() {
    return init.createCandidateSolution();
  }

  @Override
  public PermutationToBitVectorProblem split() {
    return new PermutationToBitVectorProblem(this);
  }

  /**
   * Converts a BitVector to a Permutation. Assumes that the length of the BitVector bits is
   * supportedBitVectorLength(), and behavior is undefined otherwise.
   *
   * @param bits The BitVector
   * @return A Permutation derived from the BitVector
   */
  public final Permutation toPermutation(BitVector bits) {
    Permutation p = new Permutation(permutationLength, 0);
    if (permutationLength > 1) {
      BitVector.BitIterator iter = bits.bitIterator(bitsPerElement);
      for (int remaining = permutationLength; remaining > 1; remaining--) {
        int j = permutationLength - remaining;
        int i = j + (iter.nextBitBlock() % remaining);
        if (i != j) {
          p.removeAndInsert(i, j);
        }
      }
    }
    return p;
  }

  /**
   * Computes the length of BitVectors supported by this instance.
   *
   * @return the length of BitVectors supported by this instance
   */
  public final int supportedBitVectorLength() {
    return bitsPerElement * (permutationLength - 1);
  }

  /**
   * This class implements a mapping between Permutation problems and BitVector problems, where cost
   * values are of type double. This enables using {@link BitVector BitVector} search operators to
   * solve problems defined over the space of {@link Permutation Permutation} objects. It can also
   * be used as an {@link Initializer} of BitVectors by search algorithms to ensure that the search
   * is using BitVectors of the appropriate length to represent permutations of the desired length
   * for the problem you are solving. In fact, to ensure that your search is using the correct bit
   * length, you should use this as your Initializer.
   *
   * 
The superclass, {@link PermutationToBitVectorProblem}, handles the transformation between
   * BitVectors and Permutations. See that class's documentation for the details of how a BitVector
   * is interpreted as a Permutation.
   *
   * @author Vincent A. Cicirello, https://www.cicirello.org/
   */
  public static final class DoubleCost extends PermutationToBitVectorProblem
      implements OptimizationProblem {

    private final OptimizationProblem problem;

    /**
     * Initializes the mapping between Permutation problem and BitVector problem for a specific
     * permutation length.
     *
     * @param problem The original Permutation problem.
     * @param permutationLength The length of the permutations under optimization, in number of
     *     elements. This is NOT the length of the BitVectors. For example, if the problem is the
     *     Traveling Salesperson, and if the instance has 100 cities, then you would pass 100 for
     *     this parameter.
     * @throws IllegalArgumentException if permutationLength is less than 1.
     */
    public DoubleCost(OptimizationProblem problem, int permutationLength) {
      super(permutationLength);
      this.problem = problem;
    }

    /*
     * package private for use by split
     */
    DoubleCost(DoubleCost other) {
      super(other);

      // thread safe so just copy reference
      problem = other.problem;
    }

    @Override
    public DoubleCost split() {
      return new DoubleCost(this);
    }

    @Override
    public double cost(BitVector candidate) {
      return problem.cost(toPermutation(candidate));
    }

    @Override
    public double costAsDouble(BitVector candidate) {
      return problem.costAsDouble(toPermutation(candidate));
    }

    @Override
    public boolean isMinCost(double cost) {
      return problem.isMinCost(cost);
    }

    @Override
    public double minCost() {
      return problem.minCost();
    }

    @Override
    public double value(BitVector candidate) {
      return problem.value(toPermutation(candidate));
    }
  }

  /**
   * This class implements a mapping between Permutation problems and BitVector problems, where cost
   * values are of type int. This enables using {@link BitVector BitVector} search operators to
   * solve problems defined over the space of {@link Permutation Permutation} objects. It can also
   * be used as an {@link Initializer} of BitVectors by search algorithms to ensure that the search
   * is using BitVectors of the appropriate length to represent permutations of the desired length
   * for the problem you are solving. In fact, to ensure that your search is using the correct bit
   * length, you should use this as your Initializer.
   *
   * The superclass, {@link PermutationToBitVectorProblem}, handles the transformation between
   * BitVectors and Permutations. See that class's documentation for the details of how a BitVector
   * is interpreted as a Permutation.
   *
   * @author Vincent A. Cicirello, https://www.cicirello.org/
   */
  public static final class IntegerCost extends PermutationToBitVectorProblem
      implements IntegerCostOptimizationProblem {

    private final IntegerCostOptimizationProblem problem;

    /**
     * Initializes the mapping between Permutation problem and BitVector problem for a specific
     * permutation length.
     *
     * @param problem The original Permutation problem.
     * @param permutationLength The length of the permutations under optimization, in number of
     *     elements. This is NOT the length of the BitVectors. For example, if the problem is the
     *     Traveling Salesperson, and if the instance has 100 cities, then you would pass 100 for
     *     this parameter.
     * @throws IllegalArgumentException if permutationLength is less than 1.
     */
    public IntegerCost(IntegerCostOptimizationProblem problem, int permutationLength) {
      super(permutationLength);
      this.problem = problem;
    }

    /*
     * package private for use by split
     */
    IntegerCost(IntegerCost other) {
      super(other);

      // thread safe so just copy reference
      problem = other.problem;
    }

    @Override
    public IntegerCost split() {
      return new IntegerCost(this);
    }

    @Override
    public int cost(BitVector candidate) {
      return problem.cost(toPermutation(candidate));
    }

    @Override
    public double costAsDouble(BitVector candidate) {
      return problem.costAsDouble(toPermutation(candidate));
    }

    @Override
    public boolean isMinCost(int cost) {
      return problem.isMinCost(cost);
    }

    @Override
    public int minCost() {
      return problem.minCost();
    }

    @Override
    public int value(BitVector candidate) {
      return problem.value(toPermutation(candidate));
    }
  }
}