# Set up parameters
meanA <- 1.5 # mean for class A
meanB <- 0 # mean for class B
sdA <- 1 # standard deviation for A
sdB <- 0.7 # standard deviation for B

# Create 100 random points and store them into 50 x 2 matrix
matA <- matrix(data = rnorm(n = 100, mean = meanA, sd = sdA), nrow = 50, ncol = 2)
matB <- matrix(data = rnorm(n = 100, mean = meanB, sd = sdB), nrow = 50, ncol = 2)

# Combine 2 matrix 
points <- rbind(matA, matB)

# Label A as 1, B as -1
labels <- matrix(c(rep(1, 50), rep(-1,50)))

# Now plot class A as a red point, B as a blue point
plot(points, col = ifelse(labels > 0, 2, 4))

# Splitting dataset into a train set and a test set
sample_index <- sample(100,70) # randomly pick 70 indecies

train_data <- points[sample_index,]
test_data <- points[-sample_index,]
train_label <- labels[sample_index,]
test_label <- labels[-sample_index,]

# Import class library
library(class)

# kNN Classifier with k = 1
classifier_1 <- knn(train = train_data, test = test_data, cl = train_label, k = 1)
# kNN Classifier with k = 3
classifier_3 <- knn(train = train_data, test = test_data, cl = train_label, k = 3)
# kNN Classifier with k = 5
classifier_5 <- knn(train = train_data, test = test_data, cl = train_label, k = 5)
# kNN Classifier with k = 7
classifier_7 <- knn(train = train_data, test = test_data, cl = train_label, k = 7)

# Here is a simple way to check accuracy
mean(classifier_1 == test_label)
mean(classifier_3 == test_label)
mean(classifier_5 == test_label)
mean(classifier_7 == test_label)

table(test_label, classifier_1)
table(test_label, classifier_3)
table(test_label, classifier_5)
table(test_label, classifier_7)

library(e1071)

ind <- sample(150, 150*0.8)
iris.train <- iris[ind,]
iris.test <- iris[-ind,]
fit <- naiveBayes(Species~., data=iris.train)

pred <- predict(newdata=iris.test, object = fit)
table(pred, iris.test$Species)

# import library e1071
library("e1071")

# If you are using SVM as a classifier, you need to input non-numeric value to the function
svm.classifier <- svm(train_data, as.factor(train_label))

# Now test your SVM model with test_data
svm.prediction <- predict(svm.classifier, test_data)

# Check the accuracy
sum(svm.prediction == test_label) / 30

table(test_label, svm.prediction)

summary(svm.classifier)

# Optimize SVM classifier
tune.svm(x = train_data, y = factor(train_label), cost = 10^(-1:3), gamma =c(.5:2))

# Create new classifier with different parameters
tuned_svm_classifier <- svm(train_data, as.factor(train_label), cost = 0.1, gamma = 0.5)
tuned_prediction <- predict(tuned_svm_classifier, test_data)

# Check the accuracy
sum(tuned_prediction == test_label) / 30