All Downloads are FREE. Search and download functionalities are using the official Maven repository.

org.cicirello.search.operators.bits.DefiniteBitFlipMutation Maven / Gradle / Ivy

Go to download

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.operators.bits;

import org.cicirello.math.rand.RandomIndexer;
import org.cicirello.math.rand.RandomSampler;
import org.cicirello.search.operators.IterableMutationOperator;
import org.cicirello.search.operators.MutationIterator;
import org.cicirello.search.operators.UndoableMutationOperator;
import org.cicirello.search.representations.BitVector;

/**
 * DefiniteBitFlipMutation implements a variation of Bit Flip Mutation. The form of bit flip
 * mutation commonly used in genetic algorithms (and implemented in the class {@link
 * BitFlipMutation}) is not guaranteed to change any bits during a mutation. For a metaheuristic
 * that operates on a single solution rather than a population of solutions, such as simulated
 * annealing and hill climbers, where we might use a mutation operator to generate neighbors, then
 * we will want mutation to always make some change. The DefiniteBitFlipMutation class is a
 * variation of the classic bit flip that guarantees at least 1 bit will be flipped during each
 * invocation of the mutation operator.
 *
 * 

Genetic Algorithm style Bit Flip Mutation: In a bit flip mutation, each bit is flipped with * probability M, known as the mutation rate. Flipping a bit means changing it to 0 if it is * currently a 1, or changing it to 1 if it is currently a 0. If the length of the BitVector is N, * then the expected number of bits flipped during a single mutation operation is NM. However, there * is no guarantee that any bits will be flipped during a genetic algorithm style bit flip mutation. * This behavior is fine for genetic algorithms, but may be less than desirable for other * metaheuristics, such as those that operate on a single candidate solution rather than a * population of them. * *

Definite Bit Flip Mutation: This class does not implement the genetic algorithm style bit flip * mutation. Instead, it implements a variation of it that we call Definite Bit Flip, which * guarantees that at least 1 bit will be flipped during a call to the {@link #mutate} method. * Instead of a mutation parameter, the Definite Bit Flip Mutation uses a parameter B which is an * upper bound on the number of bits that can be flipped during a single call to the {@link #mutate} * method. When the {@link #mutate} method is called, the mutation operator picks a number of bits * to flip, f, uniformly at random from the interval [1, B]. It then flips f randomly selected bits, * where all combinations of f bits are equally likely. The expected number of bits flipped during a * single call to the {@link #mutate} method is (1+B)/2. * * @author Vincent A. Cicirello, https://www.cicirello.org/ */ public final class DefiniteBitFlipMutation implements UndoableMutationOperator, IterableMutationOperator { private final int b; private int[] flipped; /** * Constructs a DefiniteBitFlipMutation operator. * * @param b The maximum number of bits to flip during a single call to the {@link #mutate} method. * The number of bits flipped during each call to {@link #mutate} method is chosen uniformly * at random from the interval [1, b]. * @throws IllegalArgumentException if b is less than 1. */ public DefiniteBitFlipMutation(int b) { if (b < 1) throw new IllegalArgumentException("b must be at least 1"); this.b = b; } /* * internal copy constructor */ private DefiniteBitFlipMutation(DefiniteBitFlipMutation other) { b = other.b; } @Override public void mutate(BitVector c) { flipped = RandomSampler.sample( c.length(), RandomIndexer.nextBiasedInt(min(b, c.length())) + 1, (int[]) null); for (int i = 0; i < flipped.length; i++) { c.flip(flipped[i]); } } @Override public void undo(BitVector c) { if (flipped != null) { for (int i = 0; i < flipped.length; i++) { c.flip(flipped[i]); } } } @Override public DefiniteBitFlipMutation split() { return new DefiniteBitFlipMutation(this); } @Override public MutationIterator iterator(BitVector c) { return new BitFlipIterator(c, min(b, c.length())); } private static int min(int x, int y) { return x < y ? x : y; } }





© 2015 - 2025 Weber Informatics LLC | Privacy Policy