Plotting minimization graph in GA

Question

The GA package maximize the fitness function in nature. I understand that for minization problem using GA, you will need to multiply your fitness with -1 in your fitness function.

##calling the ga() function
my_ga <- ga(...)

When plotting the graph (plot(my_ga)) for a minimization problem, I got a graph which shows a maximization pattern:

I tried to reverse the y-axis by the changing the ylim in plot() and I got this (which is something I wanted, except for the negative y-axis):

## explicitly defining the y-axis
x <- seq(1.1,1.4, 0.05)
plot(my_ga, ylim=rev(range(-x)))

But I don't find this approach efficient at all, is there a faster way - which to change the y-axis to positive and implicitly determining the range, to plot minimization graph using the GA package?

btw, just ignore the poor convergence in this example

[EDIT] - added source code

library(clusterSim)
library(GA)

## data
data(data_ratio)
dataset2 <- data_ratio

## fitness function
DBI2 <- function(x) {
        cl <- kmeans(dataset2, centers = dataset2[x==1, ])
        dbi <- index.DB(dataset2, cl=cl$cluster, centrotypes = "centroids")
        score <- -dbi$DB  

        return(score)  

}

k_min <- 5

## initialization of populaton
initial_population <- function(object) {
        init <- t(replicate(object@popSize, sample(rep(c(1, 0), c(k_min, object@nBits - k_min))), TRUE))
        return(init)
}

## mutation operator
my_mutation <- function(object, parent){

        pop <- parent <- as.vector(object@population[parent, ])
        difference <- abs(sum(pop) - k_min)
        ## checking condition where there are more cluster being turned on than k_min
        if(sum(pop) > k_min){
                ## determine position of 1's
                bit_position_1 <- sample(which(pop==1), difference)
                ## bit inversion
                for(i in 1:length(bit_position_1)){
                        pop[bit_position_1[i]] <- abs(pop[bit_position_1[i]] - 1)
                } 
        } else if (sum(pop) < k_min){
                ## determine position of 0's
                bit_position_0 <- sample(which(pop==0), difference)
                ## bit inversion
                for(i in 1:length(bit_position_0)){
                        pop[bit_position_0[i]] <- abs(pop[bit_position_0[i]] - 1)
                }

        } 

        return(pop)
}

my_ga<- ga(type = "binary", 
        population = initial_population, 
        fitness = DBI2, 
        selection = ga_rwSelection,
        crossover = gabin_spCrossover,
        mutation = my_mutation,
        pcrossover = 0.8,
        pmutation = 1.0,
        popSize = 100, 
        maxiter = 100,
        seed = 212,
        nBits = nrow(dataset2))


plot(my_ga)

It's easier to help you if you provide a [reproducible example](http://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example). — MrFlick, Mar 23 '17 at 15:37

score 3 · Accepted Answer · answered Mar 24 '17 at 20:23

Well, we could cheat by flipping the sign of the data in the object itself. Here's a helper function to do just that

flip_ga_y <- function(x) {
    x@summary <- x@summary*-1
    x
}

Then you can plot your sample data like

plot(flip_ga_y(my_ga), ylim=c(1.1,1.4))

Note that you will have to specify the ylim because the function makes some strong choices about the direction of the sign it seems. But this returns

Plotting minimization graph in GA

1 Answers1