CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased] - 2022-10-10

Added

• AdaptiveEvolutionaryAlgorithm class: An implementation of an evolutionary algorithm with control parameters (crossover and mutation rates) that evolve during the search.
• Option to constrain the result of a mutation on a real-valued representation to an interval (e.g., such as relevant if mutating a crossover or mutation rate since such rates constrained in [0.0, 1.0]).
• A set(int[]) method to the IntegerValued interface along with implementations in corresponding classes.
• A set(double[]) method to the RealValued interface along with implementations in corresponding classes.

Changed

• Refactored evolutionary algorithm classes to improve maintainability, as well as a few minor optimizations.
• Refactored mutation classes for real-valued representations, including:
  • Extracted an internal abstract base class for common functionality.
  • Refactored Gaussian mutation classes to use new abstract base class.
  • Refactored Cauchy mutation classes to use new abstract base class.
  • Refactored uniform mutation classes to use new abstract base class.
• Refactored several test classes based on suggestions by Sonatype Lift's new technical debt scan.

Deprecated

Removed

Fixed

Dependencies

• Bump core from 2.2.2 to 2.4.0
• Bump rho-mu from 2.4.2 to 2.5.0

CI/CD

Other

[6.0.0] - 2022-09-02

BREAKING CHANGES: This release includes breaking changes (see Removed and Changed sections).

Added

• Crossover operators for permutations:
  • Cycle Crossover (CX)
  • Order Crossover (OX)
  • Non-Wrapping Order Crossover (NWOX)
  • Uniform Order-Based Crossover (UOBX)
  • Order Crossover 2 (OX2)
  • Partially Matched Crossover (PMX)
  • Uniform Partially Matched Crossover (UPMX)
  • Precedence Preservative Crossover (PPX): two-point version
  • Precedence Preservative Crossover (PPX): uniform version
  • Edge Recombination
  • Enhanced Edge Recombination
• Additional selection operators for use in evolutionary algorithms, including:
  • Boltzmann selection with 3 temperature schedules: constant, linear cooling, exponential cooling
  • Boltzmann selection with Stochastic Universal Sampling with 3 temperature schedules: constant, linear cooling, exponential cooling

Changed

• For clarity, renamed GenerationalNANDOperatorsEvolutionaryAlgorithm to GenerationalEvolutionaryAlgorithmMutuallyExclusiveOperators (BREAKING CHANGE).

Removed

• The following classes, previously deprecated in 5.1.0:
  • ShiftedFitnessProportionalSelection, replaced by combination of FitnessShifter with FitnessProportionalSelection.
  • ShiftedStochasticUniversalSampling, replaced by combination of FitnessShifter with StochasticUniversalSampling.
  • BiasedShiftedFitnessProportionalSelection, replaced by combination of FitnessShifter with BiasedFitnessProportionalSelection.
  • BiasedShiftedStochasticUniversalSampling, replaced by combination of FitnessShifter with BiasedStochasticUniversalSampling.

Dependencies

• Bump jpt from 4.0.0 to 4.2.0
• Bump rho-mu from 2.3.2 to 2.4.2

[5.2.0] - 2022-08-01

Added

• OnePlusOneEvolutionaryAlgorithm, an implementation of a (1+1)-EA.
• OnePlusOneGeneticAlgorithm, an implementation of a (1+1)-GA, a subclass of OnePlusOneEvolutionaryAlgorithm for the special case of optimizing a BitVector representation.

[5.1.0] - 2022-07-29

Added

• SigmaScaling class, implementing sigma scaling.
• FitnessShifter class for use in combination with fitness-weighted selection operators to shift all fitnesses such that minimum fitness is 1 at the time of selection (e.g., deals with negative fitnesses, etc).

Deprecated

• ShiftedFitnessProportionalSelection, replaced by combination of FitnessShifter with FitnessProportionalSelection.
• ShiftedStochasticUniversalSampling, replaced by combination of FitnessShifter with StochasticUniversalSampling.
• BiasedShiftedFitnessProportionalSelection, replaced by combination of FitnessShifter with BiasedFitnessProportionalSelection.
• BiasedShiftedStochasticUniversalSampling, replaced by combination of FitnessShifter with BiasedStochasticUniversalSampling.

[5.0.1] - 2022-07-25

Dependencies

• Bump core from 2.1.0 to 2.2.2
• Bump rho-mu from 2.3.0 to 2.3.2

Other

• First release available via JitPack after configuring builds. In addition to Maven Central and GitHub Packages, the library can now be imported from JitPack as a fall-back option, as well as a source of snapshot artifacts built from the current default branch or specific commit hashes.

[5.0.0] - 2022-06-03

BREAKING CHANGES: This release includes breaking changes, including increasing the minimum supported Java to Java 17. See details below for other breaking changes.

Changed

• Minimum supported Java bumped to Java 17 (breaking change).
• Utilize Java 17's new RandomGenerator interface where relevant.

Dependencies

• Bumped core from 1.1.0 to 2.1.0 (breaking change).
• Bumped rho-mu from 1.2.0 to 2.3.0 (breaking change).
• Bumped jpt from 3.3.0 to 4.0.0 (breaking change).

[4.8.0] - 2022-06-02

Added

• CycleAlphaMutation: an implementation of the Cycle(alpha) form of cycle mutation.

Changed

• Refactored hierarchy and methods of stochastic sampling classes.
• Refactored exchangeBits methods and related of the BitVector class.
• Replaced calls to methods deprecated in jpt-3.2.0.

Fixed

• Added missing FunctionalInterface annotation on HeuristicBiasedStochasticSampling.BiasFunction.
• Added missing FunctionalInterface annotation on ValueBiasedStochasticSampling.BiasFunction.

Dependencies

• Bumped jpt from 3.1.1 to 3.3.0.

[4.7.0] - 2022-03-16

Added

• Factory method to LargestCommonSubgraph for creation of instances of the problem with strongly regular graphs, specifically such that the instance created consists of a pair of isomorphic Generalized Petersen Graphs.

[4.6.0] - 2022-03-14

Added

• Factory method QuadraticAssignmentProblem.createInstance for creating QAP instances from specified cost and distance matrices.
• Constructor for LargestCommonSubgraph class for creating instances from lists of the edges of the two graphs.

Changed

• Migrated all JUnit tests from JUnit 4.13.2 to JUnit Jupiter 5.8.2.

[4.5.0] - 2022-02-22

Added

• Implementation of the Largest Common Subgraph problem

Changed

• Bumped dependency jpt to 3.1.1
• Bumped dependency rho-mu to 1.2.0

[4.4.0] - 2022-02-13

Added

• Quadratic Assignment Problem

[4.3.0] - 2022-02-11

Added

• Linear Rank Selection
• Linear Rank Stochastic Universal Sampling
• Exponential Rank Selection
• Exponential Rank Stochastic Universal Sampling
• Implementation of the Bin Packing Problem, including instance generators

Changed

• Bumped dependency org.cicirello.core to 1.1.0

[4.2.1] - 2022-01-27

Fixed

• Implemented the optional method, minCost, of the OptimizationProblem and IntegerCostOptimizationProblem classes in all of the various TSP implementations to return a simple, extremely loose, lower bound of 0, to enable using the InverseCostFitnessFunction class that require a finite lower bound. The default implementation of that method otherwise returns negative infinity.

[4.2.0] - 2022-01-24

Added

• Class RandomTSPMatrix that enables generating random instances of the Traveling Salesperson Problem (TSP) as well as the Asymmetric TSP (ATSP) by generating a random distance matrix. This new class includes the option to generate a random distance matrix that enforces the triangle inequality. The library had previously introduced a TSP class in a prior release that bases the instances on points in 2D space with a user defined metric. This new RandomTSPMatrix class compliments that class by additionally providing the ability to specify arbitrary distance matrices for a TSP instance.

Changed

• Refactored existing Traveling Salesperson Problem classes (all non-breaking changes) to ease the addition of other variations.

CI/CD

• Integrated Snyk code scanning on pushes/PRs.

[4.1.0] - 2022-01-13

Added

• Transformation between Permutation problem and BitVector problem, enabling using operators for BitVectors when solving Permutation problems.

CI/CD

• Revised CI/CD workflows to deploy API website changes automatically upon each new release.

[4.0.0] - 2022-01-05

CONTAINS BREAKING CHANGES

This release contains breaking changes. See the Removed list below for details. The most significant breaking changes relate to functionality initially introduced in v3.1.0, released on 12/21/2021, and thus likely affects very few users. Other breaking changes relate to removal of previously deprecated methods.

Added

• Added GenerationalMutationOnlyEvolutionaryAlgorithm, moving the mutation-only EA functionality into this new class from the existing GenerationalEvolutionaryAlgorithm class.
• Added GenerationalNANDOperatorsEvolutionaryAlgorithm, moving the EA functionality related to mutually-exclusive genetic operators (i.e., generation variation where each child is result of either mutation, or crossover, or identical copy of a parent, but never result of both mutation and crossover) into this new class from the existing GenerationalEvolutionaryAlgorithm class.
• Enhanced all of the following Evolutionary Algorithm and Genetic Algorithm classes with the option to use elitism:
  • GenerationalEvolutionaryAlgorithm
  • GenerationalMutationOnlyEvolutionaryAlgorithm
  • GenerationalNANDOperatorsEvolutionaryAlgorithm
  • GeneticAlgorithm
  • MutationOnlyGeneticAlgorithm
• Crossover operators for IntegerVector and BoundedIntegerVector classes, including:
  • Single-point crossover
  • Two-point crossover
  • K-point crossover
  • Uniform crossover
• Crossover operators for RealVector and BoundedRealVector classes, including:
  • Single-point crossover
  • Two-point crossover
  • K-point crossover
  • Uniform crossover
• Enhancements to IntegerVector and BoundedIntegerVector, including:
  • IntegerVector.exchange method which exchanges a subsequence between two IntegerVectors.
  • BoundedIntegerVector.sameBounds method which checks if two bounded integer vectors are subject to the same min and max values.
• Enhancements to RealVector and BoundedRealVector, including:
  • RealVector.exchange method which exchanges a subsequence between two RealVectors.
  • BoundedRealVector.sameBounds method which checks if two bounded real vectors are subject to the same min and max values.
• Added constructors to TSP (traveling salesperson problem) classes to enable specifying problem instance from arrays of x and y coordinates.

Changed

• Refactored evolutionary algorithm classes to improve maintainability, and ease integration of planned future functionality.
• Refactored ProgressTracker.update(SolutionCostPair) to simplify logic.

Removed

• Removed all mutation-only EA functionality from the GenerationalEvolutionaryAlgorithm, moving that functionality into the new GenerationalMutationOnlyEvolutionaryAlgorithm. This was done for maintainability. It is a breaking-change, although it should affect minimal users as the GenerationalEvolutionaryAlgorithm was just introduced in v3.1.0. For those using it simply use the new GenerationalMutationOnlyEvolutionaryAlgorithm class where you will find all the same constructors and methods necessary for mutation-only EAs.
• Removed all mutually-exclusive genetic operator functionality from the GenerationalEvolutionaryAlgorithm, moving that functionality into the new GenerationalNANDOperatorsEvolutionaryAlgorithm. This was done for maintainability. It is a breaking-change, although it should affect minimal users as the GenerationalEvolutionaryAlgorithm was just introduced in v3.1.0. For those using it simply use the new GenerationalNANDOperatorsEvolutionaryAlgorithm class where you will find that functionality.
• All methods/constructors that were previously deprecated in earlier releases have now been removed, including:
  • The following deprecated methods of the ProgressTracker class have been removed:
    • update(int, T) in favor of using update(int, T, boolean).
    • update(double, T) in favor of using update(double, T, boolean).
    • setFoundBest() in favor of using either update(int, T, boolean) or update(double, T, boolean).
  • The following deprecated constructors of the SolutionCostPair class have been removed:
    • SolutionCostPair(T, int) in favor of using SolutionCostPair(T, int, boolean)
    • SolutionCostPair(T, double) in favor of using SolutionCostPair(T, double, boolean)

[3.1.0] - 2021-12-21

Added

• Enhancements to BitVector class, including:
  • A new method to exchange a sequence of bits between two BitVectors.
  • A new method to exchange a selection of bits, specified with a bit mask, between two BitVectors.
  • A new constructor, BitVector(length, p), that generates random bit masks given probability p of a 1-bit.
• Generational evolutionary algorithms, including the following features and functionality:
  • A generic base class, GenerationalEvolutionaryAlgorithms, with type parameter to specify type of structure we are evolving, and which supports the following EA variations:
    • Standard generational model, parents replaced by children, crossover and mutation.
    • Generational with mutation-only (no crossover).
    • Generational, parents replaced by children, with mutually exclusive crossover and mutation (i.e., no child is result of both crossover and mutation).
  • Subclasses of GenerationalEvolutionaryAlgorithm for Genetic Algorithm specific cases for convenience when optimizing BitVectors, including:
    • GeneticAlgorithm: a standard generational GA with bit flip mutation, and choice of crossover operator and selection operator.
    • MutationOnlyGeneticAlgorithm: a generational GA with only bit flip mutation, and choice of selection operator.
    • SimpleGeneticAlgorithm: a generational GA with bit flip mutation, single-point crossover, and fitness proportional selection.
  • Crossover features (mirrors the existing features of mutation operators):
    • CrossoverOperator interface for defining custom crossover operators.
    • Support for hybrid crossover operators (e.g., picking randomly from set of crossover operators).
  • Crossover operators for BitVectors, including:
    • Single-point crossover
    • Two-point crossover
    • K-point crossover
    • Uniform crossover
  • The following selection operators:
    • FitnessProportionalSelection
    • StochasticUniversalSampling
    • TournamentSelection
    • TruncationSelection
    • RandomSelection
    • BiasedFitnessProportionalSelection, which is fitness proportional selection but which enables transforming the fitness values by a bias function.
    • BiasedStochasticUniversalSampling, which is stochastic universal sampling but which enables transforming the fitness values by a bias function.
    • ShiftedFitnessProportionalSelection, which is a variation of FitnessProportionalSelection that uses transformed fitness values that are shifted so that the minimum is equal to 1, enabling safe use with negative fitness values.
    • ShiftedStochasticUniversalSampling, which is a variation of StochasticUniversalSampling that uses transformed fitness values that are shifted so that the minimum is equal to 1, enabling safe use with negative fitness values.
    • BiasedShiftedFitnessProportionalSelection, which is a variation of BiasedFitnessProportionalSelection that uses transformed fitness values that are shifted so that the minimum is equal to 1, enabling safe use with negative fitness values.
    • BiasedShiftedStochasticUniversalSampling, which is a variation of BiasedStochasticUniversalSampling that uses transformed fitness values that are shifted so that the minimum is equal to 1, enabling safe use with negative fitness values.
  • Classes to transform cost values to fitness values:
    • InverseCostFitnessFunction: Transforms optimization cost functions to fitnesses with a division.
    • NegativeCostFitnessFunction: Transforms cost to fitness with fitness(s)=-cost(s).
    • NegativeIntegerCostFitnessFunction: Transforms cost to fitness with fitness(s)=-cost(s), same as above but for integer cost problems.
• Constructive heuristics for the TSP for use by stochastic sampling algorithms, including:
  • Nearest city heuristic
  • Nearest city pair heuristic, which prefers the first city of the nearest 2-city combination
• New methods added to ProgressTracker class:
  • update(int, T, boolean)
  • update(double, T, boolean)
  • update(SolutionCostPair<T>)
• New constructors in SolutionCostPair class:
  • SolutionCostPair(T, int, boolean)
  • SolutionCostPair(T, double, boolean)

Deprecated

• The following methods of the ProgressTracker class have been deprecated:
  • update(int, T) in favor of using update(int, T, boolean).
  • update(double, T) in favor of using update(double, T, boolean).
  • setFoundBest() in favor of using either update(int, T, boolean) or update(double, T, boolean).
• The following constructors of the SolutionCostPair class have been deprecated:
  • SolutionCostPair(T, int) in favor of using SolutionCostPair(T, int, boolean)
  • SolutionCostPair(T, double) in favor of using SolutionCostPair(T, double, boolean)

CI/CD

• Updated CI/CD workflow to comment on PRs with the coverage and branches coverage percentages.

[3.0.0] - 2021-10-25

Added

• Implementation of the Traveling Salesperson Problem (class TSP), with the following features:
  • Generates random instances with cities distributed uniformly at random within a square.
  • Defaults to Euclidean distance, but also supports specifying a function for edge distance.
  • Two variations with a precomputed matrix of edge costs: floating-point costs, and integer costs, where the integer cost version by default rounds each edge cost to nearest int, but which can be customized. These use quadratic memory.
  • Two variations where edge costs are computed as needed: floating-point costs, and integer costs, where the integer cost version by default rounds each edge cost to nearest int, but which can be customized. These use linear memory.

Changed

• Beginning with release 3.0.0, the minimum supported Java version is now Java 11+.
• The randomization utilities for generating Gaussian-distributed random numbers, previously contained in the package org.cicirello.math.rand has been moved to a new library ρμ, which is a transitive dependency (via our dependency on JPT).
• Refactored GaussianMutation to improve maintainability by eliminating dependence upon a specific algorithm for generating Gaussian distributed random numbers.
• The library now uses Java modules, providing the module org.cicirello.chips_n_salsa, which includes the package org.cicirello.search and all of its very many subpackages.
  • The required dependent module org.cicirello.jpt is declared with requires transitive because the Permutation class from one of the packages of that module is a parameter and/or returned by a variety of methods such as operators that manipulate permutations. As a consequence, projects that include Chips-n-Salsa as a dependency will also include JPT and its dependencies. User's dependency manager should handle this.
• Changed the default annealing schedule in the SimulatedAnnealing class to the Self-Tuning Lam adaptive schedule of Cicirello (2021), which is implemented in class SelfTuningLam.

Other

• Reorganized directory hierarchy to the Maven default (we had been using a custom directory hierarchy for quite some time from before we switched to Maven).

[2.13.0] - 2021-09-16

Added

• The Self-Tuning Lam annealing schedule (class SelfTuningLam), as described by the paper: Vincent A. Cicirello. 2021. Self-Tuning Lam Annealing: Learning Hyperparameters While Problem Solving. Applied Sciences, 11(21), Article 9828 (November 2021). https://doi.org/10.3390/app11219828.
• Implementation of Forrester et al (2008) continuous function optimization problem.
• Implementation of Gramacy and Lee (2012) continuous function optimization problem.

Other

• Added CITATION.cff file to configure GitHub's citation link.
• Edited Zenodo metadata.

[2.12.1] - 2021-7-30

Fixed

• Corrected acceptance rate calculations in AcceptanceTracker class for cases where some runs terminated early upon finding solution with cost equal to theoretical minimum cost for problem.

[2.12.0] - 2021-7-28

Added

• Implementations of various Royal Road problems (optimization problems on bit vectors).
  • RoyalRoad class which implements Mitchell et al's original two variations (with and without stepping stones) of the Royal Road problem, but generalized to BitVectors of any length as well as other block sizes (Mitchell et al's original problem was on bit vectors of length 64 with 8-bit blocks).
  • HollandRoyalRoad class implements Holland's Royal Road function.

Changed

• Completely redesigned the website for the library, including redesigning with AMP to accelerate mobile browsing.
• Upgraded JPT dependency to JPT 2.6.1.

[2.11.1] - 2021-5-13

Changed

• Code improvements based on report from initial run of MuseDev's static code analysis tool (non-functional changes to improve code).

Fixed

• Fixed null pointer bug in stochastic sampling implementations for the case when the given heuristic doesn't use an incremental evaluation.

[2.11.0] - 2021-5-11

Added

• New mutation operators for permutations:
  • UniformScrambleMutation: Like ScrambleMutation, but scrambled elements are not necessarily contiguous.
  • UndoableUniformScrambleMutation: Exactly like UniformScrambleMutation (see above), but with support for the undo method of the UndoableMutationOperator interface.
  • TwoChangeMutation: Mutation operator on permutations that is the equivalent of the classic two change operator, assuming that the permutation represents a cyclic sequence of edges.
  • ThreeOptMutation: Mutation operator on permutations that is the equivalent of the classic 3-Opt operator, which is a combination of two-changes and three-changes, assuming that the permutation represents a cyclic sequence of edges.
  • CycleMutation: Generates a random permutation cycle.

Changed

• Updated dependency to JPT, v2.6.0.
• Minor optimization to UndoableScrambleMutation, utilizing the JPT updates.

CI/CD

• Started using CodeQL code scanning on all push/pull-request events.
• Started using Muse.dev for static analysis.
• Upgraded coverage reporting to JaCoCo 0.8.7.

[2.10.0] - 2021-3-27

Added

• Enhanced support for parallel metaheuristics, including:
  • ParallelMetaheuristic class, which enables running multiple copies of a metaheuristic, or multiple metaheuristics, in parallel using multiple threads.
  • ParallelReoptimizableMetaheuristic, which enables running multiple copies of a metaheuristic that supports the reoptimize method, or multiple such metaheuristics, in parallel using multiple threads.
• Additional artificial search landscapes, including:
  • OneMaxAckley class implements the original version of the One Max problem as described by Ackley (1985), whereas the existing OneMax class in the library implements a variation. Ackley defined the problem as maximize 10 * number of one bits.
  • Plateaus class implements the Plateaus problem, originally described by Ackley (1987), and is an optimization problem over the space of bit strings, characterized by large flat areas.
  • Mix class implements the Mix problem, originally described by Ackley (1987), and is an optimization problem over the space of bit strings, that mixes the characteristics of OneMax, TwoMax, Trap, Porcupine, and Plateau problems.
• Added the following predicate methods to BitVector: allOnes, allZeros, anyOnes, anyZeros.
• Enhancements to BitVector.BitIterator, including:
  • BitVector.BitIterator.skip() method enabling skipping over bits.
  • BitVector.BitIterator.numRemaining() method to get number of remaining bits.
  • BitVector.BitIterator.nextLargeBitBlock() for getting more than 32 bits at a time.
• RotationMutation, a mutation operator for Permutations.

Changed

• Enhanced support for parallel metaheuristics, including:
  • Refactored ParallelMultistarter class to reduce redundancy among its constructors.
  • Refactored ParallelMultistarter class to utilize the new ParallelMetaheuristic as a base class.
  • Refactored ParallelReoptimizableMultistarter class to utilize the new ParallelReoptimizableMetaheuristic as its super class.

Fixed

• Bug in definition of Ackley's Trap function (had been missing a floor due to an error in a secondary source of description of problem).

[2.9.0] - 2021-3-20

Added

• Added an implementation of the TwoMax problem.
• Added an implementation of a variation of the TwoMax problem, but with two global optima, unlike the original version which has one global and one sub-optimal local optima.
• Added an implementation of Ackley's Trap function, an artificial search landscape with one global optima, and one sub-optimal local optima, where most of the search space is within the attraction basin of the local optima.
• Added an implementation of Ackley's Porcupine function, an artificial search landscape with one global optima, and an exponential number of local optima, a very rugged landscape.

[2.8.0] - 2021-3-5

Added

• Added AcceptanceTracker, which provides a mechanism for extracting information about the behavior of an annealing schedule across a set of runs of simulated annealing.
• Added CostFunctionScaler, which enables scaling the cost function of an optimization problem by a positive constant, such as if you want to explore the effects of the range of a cost function on a stochastic local search algorithm by scaling a cost function with a known range.
• Added IntegerCostFunctionScaler, which enables scaling the cost function of an optimization problem by a positive constant, such as if you want to explore the effects of the range of a cost function on a stochastic local search algorithm by scaling a cost function with a known range. This class is just like CostFunctionScaler but for problems with integer costs.

[2.7.0] - 2021-2-25

Added

• Added HybridConstructiveHeuristic, which provides the ability to use multiple heuristics with a stochastic sampling search, where a heuristic is chosen from a set of constructive heuristics at the start of each iteration of the stochastic sampler and used for all decisions made during that iteration. The class supports the following strategies for selecting the next heuristic:
  • Heuristic uniformly at random from among the available heuristics.
  • Heuristic chosen using a round robin strategy that systematically cycles over the heuristics.
  • Heuristic chosen via a weighted random selection process.
• Added versions of a few constructive heuristics for scheduling problems that precompute heuristic values upon construction of the heuristic. Many of the scheduling heuristic implementations already do this, where it is possible to do so, provided the memory and time requirements for precomputing heuristic values is linear in the number of jobs. Some heuristics that consider setup times, however, need quadratic memory if the heuristic is precomputed prior to running the search. The existing implementations of these compute the heuristic as it is needed during the search, which means for a stochastic sampler with many iterations that the same heuristic values will be computed repeatedly. We've added versions of these that precompute the heuristic values so that if your problem is small enough to afford the quadratic memory, and if you will be running enough iterations of the stochastic sampler to afford the quadratic time of this precomputation step, then you can choose to use a precomputed version.
  • A scheduling heuristic, SmallestSetupPrecompute, which is a version of SmallestSetup, but which precomputes a table of heuristic values to avoid recomputing the same heuristic values repeatedly across multiple iterations of a stochastic sampling search.
  • A scheduling heuristic, ShortestProcessingPlusSetupTimePrecompute, which is a version of ShortestProcessingPlusSetupTime, but which precomputes a table of heuristic values to avoid recomputing the same heuristic values repeatedly across multiple iterations of a stochastic sampling search.
  • A scheduling heuristic, WeightedShortestProcessingPlusSetupTimePrecompute, which is a version of WeightedShortestProcessingPlusSetupTime, but which precomputes a table of heuristic values to avoid recomputing the same heuristic values repeatedly across multiple iterations of a stochastic sampling search.

Changed

• Dependency updated to JPT v2.4.0.

[2.6.0] - 2021-1-25

Added

• A factory method, ConstantRestartSchedule.createRestartSchedules, for creating multiple restart schedules.
• Factory methods, VariableAnnealingLength.createRestartSchedules, for creating multiple restart schedules.
• Factory methods, LubyRestarts.createRestartSchedules, for creating multiple restart schedules.

Changed

• Refactored TimedParallelMultistarter and TimedParallelReoptimizableMultistarter to eliminate redundancy in common with these two classes.
• Refactored ParallelMultistarter and ParallelReoptimizableMultistarter to eliminate redundancy in common with these two classes.
• Eliminated code redundancy between ParallelReoptimizableMultistarter and TimedParallelReoptimizableMultistarter.

Removed

• Javadoc's zipped versions of the api doc search index (and added to gitignore). The search feature on the API website will still be functional as it uses the unzipped versions as a fallback. The web server should compress these anyway. From Java 15 onward, they eliminated the generation of the zipped versions all together. We're removing them here since the api website is hosted from GitHub Pages, and git detects these as changed every time javadoc runs even if the index has not actually changed, and binary files are not efficiently stored in git.

Fixed

• Minor documentation edits to clarify details of annealing schedules (ModifiedLam, OriginalModifiedLam, and SimulatedAnnealing classes).

[2.5.2] - 2020-11-11

Fixed

• Various minor bug fixes:
  • Fixed bugs in factory methods of the uniform mutation operator, the Cauchy mutation operator, and the Gaussian mutation operator for real-valued representations related to checking for invalid parameters.
  • Fixed bug in toArray method of uniform mutation operator, the Cauchy mutation operator, and the Gaussian mutation operator in case when null passed to method.
  • Fixed bug in IntegerVectorInitializer.split().

[2.5.1] - 2020-11-5

Changed

• Refactored ParallelVariableAnnealingLength to eliminate redundant code.
• Refactored constructors of window limited permutation mutation operators to eliminate redundant code.
• Refactored BlockMoveIterator, WindowLimitedBlockMoveIterator, and BlockInterchangeIterator to remove unnecessary conditions.
• Refactored internal methods of IntegerVectorInitializer to remove unnecessary checks.

Fixed

• Various minor bug fixes:
  • Fixed bug in WeightedHybridMutation's and WeightedHybridUndoableMutation's constructors in the check for non-positive weights.
  • Fixed bug in window limited permutation mutation operators in handling of limit greater than or equal to permutation length.
  • Fixed missing parameter bounds check in LubyRestarts constructor.
  • Fixed bugs in factory methods of the uniform mutation operator for integer-valued representations related to checking for invalid parameters.
  • Fixed bugs in equals methods for bounded integer-valued representations in case when vectors contain same elements, but constrained by different bounds.
  • Fixed bug in IntegerVectorInitializer.split().

[2.5.0] - 2020-11-2

Added

• EXPET heuristic, a constructive heuristic for common duedate scheduling problems.

Changed

• Minor optimizations in scheduling problem classes (heuristics and cost functions).

Fixed

• Fixed bug in parallel searches involving handling exceptions thrown by component searches.
• Added missing parameter checks in constructors of parallel metaheuristic implementations.
• Added missing parameter bounds check in LINET scheduling heuristic implementation.
• Added test cases to improve coverage.

CI/CD

• Code coverage reporting, via JaCoCo, added to build process.

[2.4.1] - 2020-10-15

Fixed

• Bug in release workflow: attaching jars to GitHub Release.

[2.4.0] - 2020-10-15

Added

• VBSS support for exponential bias functions.

Changed

• Build process migrated from Ant to Maven.
• Improved formatting of description field in pom.xml.

Removed

• The example programs have been removed from the Chips-n-Salsa repository, and now reside in their own dedicated repository: https://github.com/cicirello/chips-n-salsa-examples
• Distribution jar files removed from repository. The distribution jars are available via GitHub Releases, GitHub Packages, and Maven Central.

[2.3.0] - 2020-10-13

Added

• Problem instance generator for Common Duedate Scheduling, based on the method used to generate the instances from the OR-Lib.
• Parser for the file format specified in OR-Lib for the Common Duedate Scheduling benchmarks.
• WLPT heuristic, a constructive heuristic for common duedate scheduling problems.
• LINET heuristic, a constructive heuristic for common duedate scheduling problems.

[2.2.0] - 2020-09-24

Added

• ModifiedLamOriginal: This class is a direct implementation of the Modified Lam annealing schedule as originally described by Swartz and Boyan. Introduced to the library to enable comparing with the optimized version already included in the library in the class ModifiedLam.
• Instance generator for static scheduling with weights and duedates.

Changed

• Upgraded dependency to JPT 2.2.0.
• Modifications to API documentation website to improve browsing on mobile web browsers.

[2.1.0] - 2020-09-18

Added

• Parser for benchmark scheduling instance data files for single machine scheduling problems with weights, duedates, and sequence-dependent setup times.

Changed

• Revised README to include instructions for importing from Maven Central
• API documentation website (https://chips-n-salsa.cicirello.org) updated to html5

[2.0.0] - 2020-09-16

First release available through Maven Central. New major release, 2.0.0, due to incompatible interface changes within the org.cicirello.search.ss package. Code that doesn't depend upon any of the stochastic sampling search algorithm implementations should be unaffected by these changes, and in those cases should be safe to simply upgrade to this new version without need to change anything in dependent code.

Added

• An interface Partial to enable generalizing the various stochastic sampling search algorithms from optimizing the space of permutations to more general types.
• A method was added to ConstructiveHeuristic for creating empty Partials of the appropriate length needed by the heuristic.
• Class HeuristicSolutionGenerator, which is a generalization of existing class HeuristicPermutationGenerator, for constructing solutions to optimization problems via constructive heuristics (HeuristicPermutationGenerator now extends this new class).
• PartialIntegerVector class to support using stochastic sampling search algorithms to generate vectors of integers.
• BoundedIntegerVector and BoundedRealVector classes added to the package org.cicirello.search.representations

Changed

• ConstructiveHeuristic and IncrementalEvaluation now have a type parameter.
• ValueBiasedStochasticSampling, HeuristicBiasedStochasticSampling, and AcceptanceBandSampling also now have a type parameter.
• All of the scheduling heuristics were modified based on interface changes.
• Renamed ConstructiveHeuristic.completePermutationLength to ConstructiveHeuristic.completeLength to be more general.
• Minor edits to the example program SchedulingWithVBSS related to other changes (added missing type parameter to construction of the ValueBiasedStochasticSampling instance to avoid compiler warning).
• Workflows and pom.xml to enable publishing to both Maven Central as well as Github Packages

[1.4.0] - 2020-08-18

Added

• .zenodo.json to provide project metadata to Zenodo during releases
• Enabled DependaBot to keep dependencies up to date.

Changed

• Refactored SimulatedAnnealing to simplify implementation.
• Minor modifications to example programs that involve the SimulatedAnnealing class to use simplified constructors.
• Refactored HBSS, VBSS, AcceptanceBandSampling, and IterativeSampling to use a common abstract base class to remove code redundancy.
• Refactored HeuristicPermutationGenerator to simplify implementation.
• Updated dependency versions.

[1.3.0] - 2020-08-06

Added

• ATCS scheduling heuristic (as described by the heuristic's authors).
• DynamicATCS scheduling heuristic (dynamically updates heuristic's parameters during schedule construction).
• An example program for an industrial scheduling problem, demonstrating using a stochastic sampling search algorithm, VBSS, as well as a hybrid of VBSS and hill climbing.
• CHANGELOG.md
• Automated sitemap.xml generation for the API website.

Changed

• Improved representation of setup times for scheduling problems with sequence-dependent setups.
• Refactored ATC and Montagne scheduling heuristics, and scheduling heuristic abstract base class for performance and style.
• Refactored hill climbing classes, introducing an abstract base class, to eliminate redundant code.

Removed

• Unnecessary maven settings.xml.

Fixed

• Fixed classpath bug in Makefile (Windows use of semicolon vs Linux, etc use of colon as delimiter). Now adjusts based on OS.

[1.2.0]

Added

• Several constructive heuristics for scheduling problems for use with stochastic sampling search algorithms.
• Heuristics include: EDD, MST, SPT, WSPT, Smallest Setup, Smallest Two-Job Setup, Montagne, Weighted COVERT, Weighted Critical Ratio, ATC.
• Heuristics also include variations of several of the above that adjust for setup times for scheduling problems with sequence-dependent setups.
• Contributing guidelines.

Changed

• Refactored scheduling heuristic classes to improve readability.
• Some code optimizations within the scheduling heuristic classes.
• Improvements to README.

[1.1.0]

Added

• Constructive heuristics for single machine scheduling problems.
• Prebuilt jars now included with release.

[1.0.2]

Changed

• Minor non-functional changes related to code style.
• Edits to JOSS paper submission.

[1.0.1]

Other

• Added a workflow to the repo to automatically publish to the GitHub Package Registry on release. This release is simply to confirm that it works and to publish the package to the GPR.

[1.0.0]

First public release

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. The library most 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. 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.