org.cicirello.search.problems.IntegerCostOptimizationProblem 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.search.SolutionCostPair;
import org.cicirello.util.Copyable;

/**
 * The IntegerCostOptimizationProblem interface provides search algorithms with a way to interact
 * with an instance of an optimization problem without the need to know the specifics of the problem
 * (e.g., traveling salesperson, bin packing, etc). It specifically concerns problems whose cost
 * function is always integer valued, such as most combinatorial optimization problems.
 *
 * Classes that implement this interface should implement the {@link #value(Copyable) value(T)}
 * method such that it returns the actual optimization objective value, and should implement the
 * {@link #cost(Copyable) cost(T)} method such that lower values are better. For a minimization
 * problem, these two methods can be implemented the same, while for a maximization problem, the
 * {@link #cost(Copyable) cost(T)} method represents a transformation from maximization to
 * minimization. This enables search algorithms to be implemented without the need to know if the
 * problem is inherently minimization or maximization. That is, a search algorithm can treat every
 * problem as minimization using the {@link #cost(Copyable) cost(T)} method. Upon completion,
 * results can then be reported in terms of the actual optimization objective function, via the
 * {@link #value(Copyable) value(T)} method.
 *
 * 
Implementers of this interface should implement the {@link #minCost minCost} method to return
 * a lower bound on the minimum cost across all possible solutions to the problem instance.
 * Implementations should be fast (preferably constant time), and need not be tight. The purpose of
 * this method is to enable a search algorithm to know if further search is futile (e.g., if it
 * actually finds a solution whose cost is equal to the bound on the minimum theoretical cost). For
 * a problem with non-negative costs, a very simple implementation might simply return 0. The
 * default implementation returns Integer.MIN_VALUE.
 *
 * @param  The type of object used to represent candidate solutions to the problem.
 * @author Vincent A. Cicirello, https://www.cicirello.org/
 */
public interface IntegerCostOptimizationProblem> extends Problem {

  /**
   * Computes the cost of a candidate solution to the problem instance. The lower the cost, the more
   * optimal the candidate solution.
   *
   * @param candidate The candidate solution to evaluate.
   * @return The cost of the candidate solution. Lower cost means better solution.
   */
  int cost(T candidate);

  /**
   * A lower bound on the minimum theoretical cost across all possible solutions to the problem
   * instance, where lower cost implies better solution. The default implementation returns
   * Integer.MIN_VALUE.
   *
   * @return A lower bound on the minimum theoretical cost of the problem instance.
   */
  default int minCost() {
    return Integer.MIN_VALUE;
  }

  /**
   * Checks if a given cost value is equal to the minimum theoretical cost across all possible
   * solutions to the problem instance, where lower cost implies better solution.
   *
   * @param cost The cost to check.
   * @return true if cost is equal to the minimum theoretical cost,
   */
  default boolean isMinCost(int cost) {
    return cost == minCost();
  }

  /**
   * Computes the value of the candidate solution within the usual constraints and interpretation of
   * the problem.
   *
   * @param candidate The candidate solution to evaluate.
   * @return The actual optimization value of the candidate solution.
   */
  int value(T candidate);

  /**
   * {@inheritDoc}
   *
   * 
The default implementation delegates work to the {@link #cost} method, which is the desired
   * behavior in most (probably all) cases. You will not likely need to override this default
   * behavior.
   */
  @Override
  default SolutionCostPair getSolutionCostPair(T candidate) {
    int c = cost(candidate);
    return new SolutionCostPair(candidate, c, isMinCost(c));
  }

  /**
   * {@inheritDoc}
   *
   * The default implementation delegates work to the {@link #cost} method. You should not need
   * to override this default behavior.
   */
  @Override
  default double costAsDouble(T candidate) {
    return cost(candidate);
  }
}