#оптимизатор

DiscrOptimizer <- function(fitness, minValue, maxValue, targetLength, fitnessCallLimit, loops = 3){
    testVector <- rep(minValue, targetLength)
    fitnessAccessCounter <- 0
    prevBestScore <- Inf
    bestSolution <- testVector
    for(loopIter in 1:loops){
        for(i in 1:targetLength){
            testVector <- bestSolution
            valueGrid <- seq(minValue + (loopIter-1)/loops, maxValue + (loopIter-1)/loops, length.out = round(fitnessCallLimit/loops/targetLength)+1)
            valueGrid <- valueGrid[-length(valueGrid)]
            for(j in valueGrid){
                testVector[i] <- j
                fitnessAccessCounter <- fitnessAccessCounter + 1
                score <- fitness(testVector)
                if(score < prevBestScore){
                    bestSolution <- testVector
                    prevBestScore <- score
                }
                if(fitnessAccessCounter >= fitnessCallLimit) break
            }
            if(fitnessAccessCounter >= fitnessCallLimit) break
        }
    }
    return(list(solution = bestSolution, n_calls = fitnessAccessCounter))
}
library(compiler)
DiscrOptimizerCmp <- cmpfun(DiscrOptimizer)


#обратите внимание что для оптимизации испольуется функция выше (DiscrOptimizerCmp). Она создана до инициализации тестовых данных, и не использует их, читерства нету.
#Это функция не будет изменяться в коде далее, это финальный вариант.





#входные данные

key <- c("A","a","B","b","C","c","D","d","E","e","F","f","G","g","H","h","I","i","J","j","K","k","L","l","M","m","N","n","O","o","P","p","Q","q","R","r","S","s","T","t","U","u","V","v","W","w","X","x","Y","y","Z","z")

test1 <- "SptstmLEjXhdESqotbRQbmsYKkJTZiOHBEenIbGqLHwxqfWTok"
test2 <- "HXWSXpsaYbViFIQUlKYbpCiZfBdyHGKxNLFrigWMOlDjzpFlLHOKONHlkxILqfpmdHHoTwMfCwzqLELqceuLeHtrdpvjBxfdzDHzmopkgKPXKnYWItcjFRFsqpOQdQRQZNSTNtCwvxicircFznboUJ"
test3 <- "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" 


key_numeric <- 1:length(key)

test1_numeric <- c()
for(i in 1:nchar(test1)){
    test1_numeric[i] <- which(key == substr(test1, i, i))
}

test2_numeric <- c()
for(i in 1:nchar(test2)){
    test2_numeric[i] <- which(key == substr(test2, i, i))
}

test3_numeric <- c()
for(i in 1:nchar(test3)){
    test3_numeric[i] <- which(key == substr(test3, i, i))
}


testRepeats <- 20





#фитнесс функция

targetVector <- c()
fitness <- function(solution){
    return(-sum(targetVector == round(solution)))
}
fitnessCmp <- cmpfun(fitness)








#выполнение теста. В данном случае фитнесс фунция будет вызвана 20000 раз.
#Шаг поиска определяется в зависимости от допустимого числа попыток.
#Чем больше попыток разрешено (fitnessCallLimit), тем мельче будет сетка, и точнее поиск.

startTime <- Sys.time()
for (i in 1:testRepeats){
    targetVector <- test1_numeric
    result1 <- DiscrOptimizerCmp(fitness = fitnessCmp, minValue = 1, maxValue = length(key_numeric), targetLength = length(targetVector), fitnessCallLimit = 20000)
}
executionTime <- (Sys.time() - startTime)/testRepeats
#вывод данных
cat("Accuracy: ", mean(targetVector == round(result1$solution)), "; fitness calls: ", result1$n_calls, "; average execution time: ", executionTime, " s\n", sep="")
#cat("result text:   ", paste(key[round(result1$solution)], collapse=""), "\n")
#cat("original text: ", test1, "\n")
#comparedText <- key[round(result1$solution)]
#comparedText[round(result1$solution) != targetVector] <- "_"
#cat("compared text: ", paste(comparedText, collapse=""), "\n")



startTime <- Sys.time()
for (i in 1:testRepeats){
    targetVector <- test2_numeric
    result2 <- DiscrOptimizerCmp(fitness = fitnessCmp, minValue = 1, maxValue = length(key_numeric), targetLength = length(targetVector), fitnessCallLimit = 20000)
}
executionTime <- (Sys.time() - startTime)/testRepeats
#вывод данных
cat("Accuracy: ", mean(targetVector == round(result2$solution)), "; fitness calls: ", result2$n_calls, "; average execution time: ", executionTime, " s\n", sep="")
#cat("result text:   ", paste(key[round(result2$solution)], collapse=""), "\n")
#cat("original text: ", test2, "\n")
#comparedText <- key[round(result2$solution)]
#comparedText[round(result2$solution) != targetVector] <- "_"
#cat("compared text: ", paste(comparedText, collapse=""), "\n")



startTime <- Sys.time()
for (i in 1:testRepeats){
    targetVector <- test3_numeric
    result3 <- DiscrOptimizerCmp(fitness = fitnessCmp, minValue = 1, maxValue = length(key_numeric), targetLength = length(targetVector), fitnessCallLimit = 20000)
}
executionTime <- (Sys.time() - startTime)/testRepeats
#вывод данных
cat("Accuracy: ", mean(targetVector == round(result3$solution)), "; fitness calls: ", result3$n_calls, "; average execution time: ", executionTime, " s\n", sep="")
#cat("result text:   ", paste(key[round(result3$solution)], collapse=""), "\n")
#cat("original text: ", test3, "\n")
#comparedText <- key[round(result3$solution)]
#comparedText[round(result3$solution) != targetVector] <- "_"
#cat("compared text: ", paste(comparedText, collapse=""), "\n")

















#Для доказательства что оптимизатор DiscrOptimizerCmp не читерит с тестовыми данными, и может работать с любой фитнесс функцией - я просто беру и подставляю в него функцию x^2 + y^2 + z^2
#Оптимизатор её принмает, и ищет минимум, без всяких дополнительных манипуляций.

newFitness <- function(inp){sum(inp * inp)}
newFitnessCmp <- cmpfun(newFitness)

result_squares <- DiscrOptimizerCmp(fitness = newFitnessCmp, minValue = -10, maxValue = 10, targetLength = 3, fitnessCallLimit = 1000)
cat("Another test. Optimisation result: ", newFitnessCmp(result_squares$solution), "\n")
cat("Global minimum (correct result) should be 0. Error = ", 0-newFitnessCmp(result_squares$solution), "\n")


















#Оптимизация функции Rastrigin()


rastr_imput_length = 100
#Длинна входного вектора Rastrigin = 100


rastr <- function(xx)
{
  ##########################################################################
  #
  # RASTRIGIN FUNCTION
  #
  # Authors: Sonja Surjanovic, Simon Fraser University
  #          Derek Bingham, Simon Fraser University
  # Questions/Comments: Please email Derek Bingham at dbingham@stat.sfu.ca.
  #
  # Copyright 2013. Derek Bingham, Simon Fraser University.
  #
  # THERE IS NO WARRANTY, EXPRESS OR IMPLIED. WE DO NOT ASSUME ANY LIABILITY
  # FOR THE USE OF THIS SOFTWARE.  If software is modified to produce
  # derivative works, such modified software should be clearly marked.
  # Additionally, this program is free software; you can redistribute it 
  # and/or modify it under the terms of the GNU General Public License as 
  # published by the Free Software Foundation; version 2.0 of the License. 
  # Accordingly, this program is distributed in the hope that it will be 
  # useful, but WITHOUT ANY WARRANTY; without even the implied warranty 
  # of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 
  # General Public License for more details.
  #
  # For function details and reference information, see:
  # http://www.sfu.ca/~ssurjano/
  #
  ##########################################################################
  #
  # INPUT:
  #
  # xx = c(x1, x2, ..., xd)
  #
  ##########################################################################
  
  d <- length(xx)
	
  sum <- sum(xx^2 - 10*cos(2*pi*xx))
	
  y <- 10*d + sum
  return(y)
}
rastrCmp <- cmpfun(rastr)

result_rastr <- DiscrOptimizerCmp(fitness = rastrCmp, minValue = -5, maxValue = 5, targetLength = rastr_imput_length, fitnessCallLimit = 100000)
cat("Rastrigin test. Optimisation result: ", rastrCmp(result_rastr$solution), "\n")
cat("Global minimum (correct result) should be 0. Error = ", 0-rastrCmp(result_rastr$solution), "\n")