$5.$ For problem $4,$ instead of implementing validation set approach, proceed to use leave$-$ one$-$out cross$-$validation $($function knn$.$cv$()).$ Run it for K $=1,3,10$ and compare the resulting CV errors. Use all observations of Auto data set for relevant pre$-$ dictors, not just the $$training subset$($as we are not doing any train$/$test subdivision here$).$

$($a$)$ Run set.seed$(1)$ command prior to each cv$.$knn$()$ call, for uniformity of the re$-$ sults.

$($b$)$ First do it for data in its original form $($as you had in $5($d$),$ no scaling$).$ What are the test errors? Which method wins? Show the code for just one of K values $($e$.$g$.$ K $=3).$

$($c$)$ Then do it for scaled data $($as you had in $5($e$)).$ What are the test errors? Which method wins? Show the code for just one of K values $($e$.$g$.$ K $=3)$

$($d$)$ Which results should we trust more $-$ validation set approach from problem $5?$ Or CV results here? Why?