HW1_backup

# Homework 1 notes

check knn vs. true classification error for 2.2.3
3.1, 3.2

may have to refigure out code to extact classification accuracy
rather than just reporting MAE and MSE

also notice that MAE is the same for k = (58, 30, 20, 10)

check documentation to see what the output of train.kknn is
 - minimal mean absolute error
 - minimal mean squared error





 # set up list of possible K values from 1 to 100 by 3
possible_k <- as.list(seq(from = 1, to = 100, by = 3))

# set up a blank list to put accuracy values into
test_accuracy <- list()

# fit a model for each possible value of K and extract the accuracy from each model
for (i in seq_along(possible_k)) {

        k = possible_k[[i]]

        knn_fit <- kknn(
                response_y ~ .,
                train = train,
                test = test,
                # valid = validation$response_y,
                k = k,
                scale = T)

        fitted <- fitted(knn_fit) %>%
             as_tibble() %>%
             mutate(value = ifelse(value > .5, 1, 0)) %>%
             cbind(., test$response_y) %>%
             mutate(acc = value == test$response_y)

        test_accuracy[[i]] <- mean(fitted$acc)

}

# put the K and test accuracy lists into dataframes
k_df <- reduce(possible_k, rbind.data.frame)
test_acc_df <- reduce(test_accuracy, rbind.data.frame)

# find the best K associated with the highest accuracy
(performance_test <- cbind(k_df, test_acc_df) %>%
        rename(
                'knn_error' = !!names(.[2]),
                'knn_k' =  !!names(.[1])
        ) %>%
        filter(knn_error == max(knn_error)) %>%
        arrange(., knn_k) %>%
        .[1,])