R/benchmark.R

In jhoffjann/nmmso_benchmark: CEC Benchmarking Tool

library(devtools)
library(nmmso.R)
source("./R/cec_2015_problem_data.R")
source("./R/niching_funcs.R")
source("./R/count_goptima.R")

# the accuracies
acc = c(0.1, 0.01, 0.001, 0.0001, 0.00001)
niching_funcs = c(five_uneven_peak_trap, equal_maxima, uneven_decreasing_maxima,
                  himmelblau, six_hump_camel_back, shubert, vincent, shubert, vincent,
                  modified_rastrigin_all, CF1, CF2, CF3, CF3, CF4, CF3, CF4, CF3, CF4, CF4)

#' @title benchmark
#' @description Function that calls the NMMSO algorithm with the 20 functions to be evaluated with.
#'
#' @param index Number of the function to be used.
#' @param seed Random seed for the evaluation.
#'
#' @export
benchmark <- function(index, seed){
    # set new seed
    set.seed(seed)
    # initialize all basic stuff for tracking the state
    eval_list = c()
    c_list = c()
    pop_size = c()
    conv_evals = c()
    count = c(0,0,0,0,0)
    
    
    # those are the initializations that normally would happen in nmmso.R
    state = list()
    evaluations_after = 0
    mode_loc_after = list()
    mode_y_after = list()
    nmmso_state = list()
    evaluations_after = 0
    iteration = 0
    
    # main loop of the program
    while(evaluations_after < gens[[index]] && count[5] != nopt[index]){
      	iteration = iteration + 1
    	old_count = count
    	count = c(0,0,0,0,0)
    	result <- NMMSO_iterative(swarm_size = as.numeric(10*length(mx[[index]])), problem_function = niching_funcs[[index]], max_evaluations = gens[[index]], mn = as.numeric(mn[[index]]), mx = as.numeric(mx[[index]]), evaluations = evaluations_after, nmmso_state = nmmso_state)
    	mode_loc_after = result$mode_loc
    	mode_y_after = result$mode_y
    	evaluations_after = result$evaluations
    	nmmso_state = result$nmmso_state
    	best = which(mode_y_after > max(mode_y_after)-0.1)

    	for(i in 1:5){
    		count[i] = count_goptima(as.matrix(mode_loc_after[best, , drop=FALSE]), index, acc[i])$count
    	}
    	
    	eval_list = rbind(eval_list, as.matrix(evaluations_after))
    	c_list = rbind(c_list, rbind(count))
    	pop_size = c(pop_size, length(mode_y_after))
    	write(c(iteration, length(nmmso_state$swarms)-1), file = paste("./output/", index, "_", seed, ".txt", sep = ""), append = TRUE)
    }
    
    # count actually found optima
    if(evaluations_after > gens[index]){
    	c_list = c_list[-nrow(c_list), ]
    	count = old_count
    }
    
    # check at which point the optimum for the different accuracy levels was found
    for(j in 1:5){
    	I = which(c_list[,j] == nopt[index])
    	if(length(I) != 0)
    		conv_evals[j] = eval_list[I[1]]
    	else
    		conv_evals[j] = gens[index] + 1
    }
    
    # write the results down.
    write(conv_evals, paste("./output/", index, "_" ,seed,"_output.txt", sep = ""))
    write.table(c_list, paste("./output/", index, "_", seed, "_clist.txt", sep = ""), col.names = FALSE, row.names = FALSE)
  }