forega (version 1.0.3)

forecasting_ga: Floating-Point Genetic Algorithms with Statistical Forecast Based Inheritance Operator

Description

This function performs a floating-point genetic algorithm search with a statistical forecasting operator that generates offspring which probably will be generated in future generations. Use of this operator makes floating-point genetic algorithms hybrid, that is, forecasting future generations mimics a local search.

Usage

forecasting_ga(evalFunc, chsize, minv, maxv, crossprob, mutationprob, 
               forecastprob, elitism, popsize, maxiter, MinimumForecastLength, 
               ForecastFunction)

Arguments

evalFunc

Function to be minimized

chsize

Chromosome size

minv

Vector of lower bounds of decision variables (genes)

maxv

Vector of upper bounds of decision variables (genes)

crossprob

Crossover probability

mutationprob

Mutation probability

forecastprob

Probability of forecasting operator applied on variables (genes)

elitism

Number of chromosomes that are directly copied into the next generations

popsize

Population size

maxiter

Maximum number of iterations (generations)

MinimumForecastLength

Minimum number of ancestors of a chromosome before applying statistical forecast

ForecastFunction

R function to be used in forecasting. ForecastArima is implemented as R code and any other forecast functions can be implemented instead

Value

Returns the population at last generation ordered by cost (fitness) values. First element is the best.

Details

EvalFunc is always encapsulates a minimization problem. Genetic operators satisfy the boundary constraints. Since the current forecast method is ARIMA, the parameter ForecastFunction must be manually set to ForecastArima if there is no any other function at hand. A new user-defined forecast function can easly be implemented which is based on the other well-known forecast methods.

Examples

# NOT RUN {
# Required package forega
require("forega")

set.seed(12345)

# This function has the global minimum at x_1 = pi and x_2 = exp(1)
f <- function (x){
	return( (x[1]-pi)^2 + (x[2]-2.71828)^2 )
}

# Performing a floating-point genetic algorithm search with forecast probability of 0.10
res <- forecasting_ga(evalFunc=f, chsize=2, minv=rep(-10.0,2),
                      maxv=rep(10.0,2), crossprob=0.80, mutationprob=0.01, 
                      popsize=100, maxiter=1000, MinimumForecastLength=20, 
                      ForecastFunction=ForecastArima, elitism=2, forecastprob=0.01)

# Show the first chromosome of the returned population matrix
print(res[1,])


# }