if (!requireNamespace("doBy", quietly = TRUE)) {
  install.packages("doBy")
}
library(doBy)

data(wine)
write.csv(wine,"wine.csv", row.names = FALSE) # simply to export it as a csv locally


# keep the original colnames
orig_colnames <- colnames(wine)
colnames(wine) <- c("Class", paste0("Feature_", 1:13))
head(wine)

# Visualize pairwise relationships
pairs(wine[ ,2:5], col=wine$Class, main="Pairwise Plot of Selected Features")
# Summary
summary(wine)


# Split data
set.seed(123)
library(caret)
train_idx <- createDataPartition(wine$Class, p=0.7, list=FALSE)
train <- wine[train_idx,]
test <- wine[-train_idx,]

# k-NN
library(class)
knn_pred <- knn(train[ ,2:14], test[ ,2:14], train$Class, k=5)
cat("k-NN Accuracy:", mean(knn_pred == test$Class), "\n")

# Decision Tree
library(rpart)
tree_model <- rpart(Class ~ ., data=train)
library(rpart.plot)
rpart.plot(tree_model)
tree_pred <- predict(tree_model, test, type="class")
cat("Decision Tree Accuracy:", mean(tree_pred == test$Class), "\n")

# Perceptron (binary: class 1 vs. classes 2 & 3)
library(neuralnet)
wine$BinaryClass <- ifelse(wine$Class == 1, 1, 0)
train_bin <- wine[train_idx,]
test_bin <- wine[-train_idx,]
perceptron_model <- neuralnet(BinaryClass ~ Feature_1 + Feature_2 + Feature_3 + Feature_4 + Feature_5 + Feature_6 + Feature_7 + Feature_8 + Feature_9 + Feature_10 + Feature_11 + Feature_12 + Feature_13,
                              data=train_bin, hidden=0, linear.output=FALSE)
preds <- compute(perceptron_model, test_bin[ ,2:14])$net.result
pred_class <- ifelse(preds > 0.5, 1, 0)
accuracy_perceptron <- mean(pred_class == test_bin$BinaryClass)
cat("Perceptron Accuracy (class 1 vs. others):", accuracy_perceptron, "\n")


## Validation
# Cross-validation for k-NN and decision tree
train_control <- trainControl(method="cv", number=5)

knn_cv <- train(as.factor(Class) ~ ., data=wine, method="knn", tuneGrid=data.frame(k=5),
                trControl=train_control)
print(knn_cv)

dt_cv <- train(as.factor(Class) ~ ., data=wine, method="rpart", trControl=train_control)
print(dt_cv)

# access the sd of the model's performance
knn_cv$results$Accuracy+c(-1,1)*1.96*knn_cv$results$AccuracySD



## clustering

set.seed(123)  # For reproducibility
# Apply k-means clustering with k=3 clusters
kmeans_res <- kmeans(wine[ ,2:14], centers=3)

# Visualize the clustering results using two features
plot(wine$Feature_1, wine$Feature_2, col=kmeans_res$cluster,
     main="k-means Clustering of Wine Data",
     xlab="Feature 1", ylab="Feature 2",
     pch=19)
# Optionally, add true class labels for comparison
points(wine$Feature_1, wine$Feature_2, col=as.numeric(wine$Class), pch=1)
legend("topright", legend=c("Cluster 1", "Cluster 2", "Cluster 3", "True Class 1", "True Class 2", "True Class 3"),
       col=c(1,2,3,1,2,3), pch=c(19,19,19,1,1,1))