rm(list=ls())

library(class) # imports knn
library(MLmetrics) # imports accuracy

glass = data(Glass, package = 'mlbench')
training = Glass[c(1:50, 91:146), -10]
trainingType = factor(Glass[c(1:50, 91:146), 10])
testing = Glass[51:90, -10]
testingType = factor(Glass[51:90, 10])

# cor(training)
pca_model <- prcomp(training, center = TRUE, scale = TRUE)
eigenvectors = pca_model$rotation
eigenvalues = pca_model$sdev^2

# Plot the variance percentage for each component
barplot(eigenvalues / sum(eigenvalues))

# Q1
eigenvalues[1]/sum(eigenvalues)

# Q2 Info loss = percentage of not included eigenvalues
sum(eigenvalues[-c(1:4)])/sum(eigenvalues)

# Q3
model = knn(training, testing, trainingType, k = 3)
Accuracy(model, testingType)

# Q4 same model - incorrect? elearning = 0.8
Recall(model, testingType, '2')

max_acc = 0
max_i = 1
for (i in 1:9) {
  pcs_training = as.data.frame(predict(pca_model, training)[, 1:i])
  pcs_testing = as.data.frame(predict(pca_model, testing)[, 1:i])
  model = knn(pcs_training, pcs_testing, trainingType, k = 3)
#  print(Accuracy(model, testingType))
 # print(i)
  acc = Accuracy(model, testingType)
  if (acc > max_acc){
    max_acc = acc
    max_i = i
  }
}

# Q5
max_acc
max_i