register_GA: Register a Genetic Algorithm Seriation Method

Description

Usage

register_GA()

Arguments

Details

Registers the method "GA" for seriate. This method can be used to optimize any criterion in package seriation. control for seriate with method "GA" accepts the following parameters:

"criterion": criterion to optimize. Default: BAR
"suggestions": suggestions to warm start the GA. NA means no warm start. Default: TSP, QAP_LS and Spectral.
"selection": Selection operator (see GA). Default: non-linear rank selection
"crossover": Crossover operator (see GA). Default: ordered crossover (OX)
"mutation": Mutation operator (see GA). Default: a mixture of the simple insertion (80% chance) and simple inversion (20% chance) operators.
"pmutation": proability for permutations. Default: .5
"pcrossover": probability for crossover. Default: .2
"popsize": the population size. Default: 100
"maxiter": maximum number of generations. Default: 1000
"run": stop after run generations without improvement. Default: 50
"parallel": use multiple cores? Default: TRUE
"verbose": Report progress? Default: TRUE

The GA uses by default the ordered cross-over (OX) operator. For mutation, the GA uses a mixture of simple insertion and simple inversion operators. This mixed operator is created using seriation::gaperm_mixedMutation(ismProb = .8), where ismProb is the probability that the simple insertion mutation operator is used. See package GA for a description of other available cross-over and mutation operators for permutaitons. The appropriate operator functions in GA start with gaperm_.

We warm start the GA using "suggestions" given by several heuristics. Set "suggestions" to NA to start with a purely random initial population.

Note: Package GA needs to be installed.

References

Luca Scrucca (2013): GA: A Package for Genetic Algorithms in R. Journal of Statistical Software, 53(4), 1--37. URL http://www.jstatsoft.org/v53/i04/.

Examples

Run this code

## Not run: ------------------------------------
# register_GA()
# list_seriation_methods("dist")
# 
# d <- dist(random.robinson(50, pre=TRUE, noise=.1))
# 
# ## use default settings: Banded AR form
# o <- seriate(d, "GA")
# pimage(d, o)
# 
# ## optimize for linear sertiation criterion (LS)
# o <- seriate(d, "GA", control = list(criterion = "LS"))
# pimage(d, o)
# 
# ## no warm start
# o <- seriate(d, "GA", control = list(criterion = "LS", suggestions = NA))
# pimage(d, o)
## ---------------------------------------------