Genetic Algorithms

Flexible general-purpose toolbox implementing genetic algorithms (GAs) for stochastic optimisation. Binary, real-valued, and permutation representations are available to optimize a fitness function, i.e. a function provided by users depending on their objective function. Several genetic operators are available and can be combined to explore the best settings for the current task. Furthermore, users can define new genetic operators and easily evaluate their performances. Local search using general-purpose optimisation algorithms can be applied stochastically to exploit interesting regions. GAs can be run sequentially or in parallel, using an explicit master-slave parallelisation or a coarse-grain islands approach.

An R package for stochastic optimisation using Genetic Algorithms.

The GA package provides a flexible general-purpose set of tools for implementing genetic algorithms search in both the continuous and discrete case, whether constrained or not. Users can easily define their own objective function depending on the problem at hand. Several genetic operators are available and can be combined to explore the best settings for the current task. Furthermore, users can define new genetic operators and easily evaluate their performances. Local search using general-purpose optimisation algorithms can be applied stochastically to exploit interesting regions. GAs can be run sequentially or in parallel, using an explicit master-slave parallelisation or a coarse-grain islands approach.

Installation

You can install the released version of GA from CRAN:

install.packages("GA")

or the development version from GitHub:

# install.packages("devtools")
devtools::install_github("luca-scr/GA", build_vignettes = TRUE)

Usage

Usage of the main functions and several examples are included in the papers shown in the references section below.

For an intro see the vignette A quick tour of GA, which is available as

vignette("GA")

The vignette is also available in the Get Started section on the GitHub web page of the package at http://luca-scr.github.io/GA/.

References

Scrucca, L. (2013) GA: A Package for Genetic Algorithms in R. Journal of Statistical Software, 53(4), 1-37. https://www.jstatsoft.org/article/view/v053i0

Scrucca, L. (2017) On some extensions to GA package: hybrid optimisation, parallelisation and islands evolution. The R Journal, 9(1), 187–206. https://journal.r-project.org/archive/2017/RJ-2017-008

News

GA 3.2 (2019-01)

Added website using pkgdown.
Added a section to vignette on integer optimisation.
Added function de() implementing Differential Evolution based on the description in Simon (2013) Evolutionary Optimization Algorithms.
Bug fix in C++ code calling pow() function.
Bug fix in gails() when suggestions are provided.
Long outputs in summary() function calls are shortened.
Fix a bug in gaperm_pbxCrossover().

GA 3.1.1 (2018-05)

Bug fix in C++ code calling pow() function.

GA 3.1 (2018-05)

Genetic operators available in C++ using Rcpp package.
Add parameter "useRcpp" in gaControl() to control if the C++ implementation of genetic operators should be used. By default is set to TRUE.
Renamed input parameters min and max to lower and upper. The old nomenclature is still accepted but deprecated.
Added stopParallel() function to stop a cluster if parallel = TRUE, including registerDoSEQ() to avoid problems if foreach loop is used after.
A parallel cluster is automatically stopped unless a registered parallel back end is provided as argument to parallel argument in the ga() or gaisl() function call.
Bug fix in gaperm_Population() when min > 1. Thanks to Romero Barata.
Update "A quick tour of GA" vignette.
Add README.Rmd that is shown in GitHub and CRAN.
Create a sticker for the GA package.

GA 3.0.2 (2016-06)

In interactive sessions use different monitor functions depending on whether or not is an RStudio session. This is due to different capabilities of the several consoles from which R can be run.

GA 3.0.1 (2016-05)

doRNG package moved from Depends to Suggests due to the unavailability of the binary version of the package on Windows. As a consequence, GA package can be installed even without doRNG being installed, but in this case results are not reproducible by setting the seed if GAs are executed in parallel.

GA 3.0 (2016-05)

Added option to provide a cluster for parallelisation.
Added GA-hybrid using optim() for local search.
Added GA Islands model in gaisl().
Rewrite of gaMonitor() to clear the previous output before printing the info about the current iteration. Old version is available in gaMonitor2(). Same behaviour for gaislMonitor() and gaislMonitor2().

GA 2.3 (2015-07)

Added documentation vignettes.

GA 2.2 (2014-10)

Slight modification to plot.ga() method.
Included link in main ga documentation to genetic operators.
Added ga_pmutation() function to allow GAs having variable mutation probability.
pdf files (previously included as vignettes) moved to inst/doc with corresponding index.html.

GA 2.1 (2014-05)

.printShortMatrix() is a function to print part of rows/columns of a matrix
print.summary.ga() accept arguments to be passed to .printShortMatrix(). This allow to shorten the printed output in case of large dimensions for the matrices containing the suggestions and the final solutions.
plot.ga() now shadowed the area between the max and the median of fitness values at each iteration. Changed to pch from 17 to 1 for means.
Function gaSummary() is embedded in the code, so it cannot be defined by the user.
Added function ga_pmutation() for computing variable mutation probability.
Modified main function ga() to allow for pmutation to be a function. This enables the use of variable mutation rate.
Export plot.ga class in NAMESPACE and used as S4 method.
Computing summaries at each step is done by function gaSummary(), but it may be defined by the user.
Export summary.ga class in NAMESPACE and used as S4 method.
Default for suggestions argument in ga() is set to NULL.
Fixed a bug in gabin_uCrossover().
Added description of slots in the help page for ga-class.
Bug fix on setting maxfitness.
Bug fix on passing seed argument when "snow" parallel is used.
Add explicitly stop clusters if parallel is used.
Parallel computing improving set up of clusters.