setwd(".")

options(stringsAsFactors = FALSE)

#library("clusterSim")

library("e1071")

library("PRROC")

tau = 0.5

source("./confusion_matrix_rates.r")

dataFileName <- "../data/LungCancerDataset_AllRecords_NORM_27reduced_features.csv"

cat("dataFileName = ", dataFileName, "\n", sep="")

cancer_data_norm <- read.csv(file=dataFileName,head=TRUE,sep=",",stringsAsFactors=FALSE)

cancer_data_norm <- cancer_data_norm[sample(nrow(cancer_data_norm)),] # shuffle the rows

totalElements <- dim(cancer_data_norm)[1]

subsets_size <- 4000

if (subsets_size != totalElements) {

    cat("!!! ATTENTION: We are running the method on a subset of the original dataset, \n", sep="")

    cat("!!! containing only ", subsets_size, " elements \n", sep="")

    cat("!!! instead of ", totalElements, " elements \n", sep="")

}

cancer_data_norm <- cancer_data_norm[1:subsets_size, ]

dataset_dim_retriever(cancer_data_norm)

imbalance_retriever(cancer_data_norm$Metastasis)

target_index <- dim(cancer_data_norm)[2]

training_set_perce <- 60

cat("training_set_perce = ", training_set_perce, "% \n", sep="")

validation_set_perce <- 20

cat("validation_set_perce = ", validation_set_perce, "% \n", sep="")

test_set_perce <- 100 - training_set_perce - validation_set_perce

cat("test_set_perce = ", test_set_perce, "% \n", sep="")

# the training set is the first 60% of the whole dataset

training_set_first_index <- 1 # NEW

training_set_last_index <- round(dim(cancer_data_norm)[1]*training_set_perce/100) # NEW

# the validation set is the following 20% of the whole dataset

validation_set_first_index <- round(dim(cancer_data_norm)[1]*training_set_perce/100)+1 # NEW

validation_set_last_index <- round(dim(cancer_data_norm)[1]*(training_set_perce+validation_set_perce)/100) # NEW

# the test set is the last 20% of the whole dataset

test_set_first_index <- round(dim(cancer_data_norm)[1]*(training_set_perce+validation_set_perce)/100)+1 # NEW

test_set_last_index <- dim(cancer_data_norm)[1] # NEW

cat("[Creating the subsets for the values]\n")

cancer_data_train <- cancer_data_norm[training_set_first_index:training_set_last_index, 1:(target_index-1)] # NEW

cancer_data_validation <- cancer_data_norm[validation_set_first_index:validation_set_last_index, 1:(target_index-1)] # NEW

cancer_data_test <- cancer_data_norm[test_set_first_index:test_set_last_index, 1:(target_index-1)] # NEW

cat("[Creating the subsets for the labels \"1\"-\"0\"]\n")

cancer_data_train_labels <- cancer_data_norm[training_set_first_index:training_set_last_index, target_index] # NEW

cancer_data_validation_labels <- cancer_data_norm[validation_set_first_index:validation_set_last_index, target_index] # NEW

cancer_data_test_labels <- cancer_data_norm[test_set_first_index:test_set_last_index, target_index]   # NEW

library(class)

library(gmodels)

# The k value must be lower than the size of the training set

maxK <- 10 #NEW

mcc_array <- character(length(maxK))

# NEW PART:

c_array = c(0.001, 0.01, 0.1, 1, 10)

mccCounter = 1

cat("\n[Optimization of the hyper-parameter C start]\n")

# optimizaion loop

for(thisC in c_array)

{

  # apply k-NN with the current K value

  # train on the training set, evaluate in the validation set by computing the MCC

  # save the MCC corresponding to the current K value

  cat("[Training the SVM model (with C=",thisC,") on training set & applying the SVM model to validation set]\n", sep="")

  svm_model <- svm(cancer_data_train_labels ~ ., cost=thisC, data=cancer_data_train, method = "C-classification", kernel = "linear")

  cancer_data_validation_PRED <- predict(svm_model, cancer_data_validation)

  cancer_data_validation_pred_binary <- as.numeric (cancer_data_validation_PRED)  

  cancer_data_validation_pred_binary[cancer_data_validation_pred_binary>=tau]<-1

  cancer_data_validation_pred_binary[cancer_data_validation_pred_binary<tau]<-0

#   # cancer_data_validation_pred_binary

#    fg_test <- cancer_data_validation_PRED[cancer_data_validation_labels==1]

#    bg_test <- cancer_data_validation_PRED[cancer_data_validation_labels==0]

# 

#    pr_curve_val <- pr.curve(scores.class0 = fg_test, scores.class1 = bg_test, curve = F)

#    # plot(pr_curve_test)

#    # print(pr_curve_val)

# 

#    roc_curve_val  <- roc.curve(scores.class0 = fg_test, scores.class1 = bg_test, curve = F)

#    # plot(pr_curve_test)

  # print(roc_curve_val)

  mcc_outcome <- mcc(cancer_data_validation_labels, cancer_data_validation_pred_binary)

  cat("When C=",thisC,", the MCC value is ",mcc_outcome, "\t (worst possible: -1; best possible: +1)\n", sep="")

  mcc_array[mccCounter] <- mcc_outcome

  mccCounter = mccCounter + 1

}

# select the k corresponding to the highest MCC and call it k_best

bestMCC <- max(mcc_array)

bestCindex <- match(bestMCC, mcc_array)

cat("\nThe best C value is ", c_array[bestCindex],", corresponding to MCC=", mcc_array[bestCindex],"\n", sep="")

cat("[Optimization end]\n\n")

cat("\n @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ \n")

# apply k-NN with k_best to the test set

cat("\n[Training the SVM model (with the OPTIMIZED hyper-parameter C=",c_array[bestCindex],") on training set & applying the SVM to the test set]\n", sep="")

#cancer_data_test_pred <- knn(train = cancer_data_train, test = cancer_data_test, cl = cancer_data_train_labels, k=bestK)

svm_model_new <- svm(cancer_data_train_labels ~ ., cost=c_array[bestCindex], data=cancer_data_train, method = "C-classification", kernel = "linear")

cancer_data_test_pred <- predict(svm_model_new, cancer_data_test)

confusion_matrix_rates(cancer_data_test_labels, cancer_data_test_pred, "@@@ Test set @@@")