$---$

title: "Finish the Incomplete Code"

output: html$\_$notebook

$---$

Libraries:

$```\{$r$\}$

library$($ISLR$2)$

library$($tree$)$

library$($tidyverse$)$

library$($caret$)$

library$($dplyr$)$

library$($factoextra$)$

$```$

First,set your seed for reproducibility:

$```\{$r$\}$

# set your seed to $0412$

set.seed$(0412)$

head$($Auto$)$

$```$

We will be using the Auto dataset for the ISLR$2$ package

Read in the data and check $10$ random observations

$```\{$r$\}$

data$($Auto$)$

sample$\_$n$($Auto$,10)$

$```$

Are there any missing values?:

$```\{$r$\}$

#Check for missing values:

is$.$na$($Auto$)$

$```$

What are the dimensions of your data?:

$```\{$r$\}$

#Check and report dimensions

dim$($Auto$)$

$```$

# $1$:Decision Trees

Regression Tree: Our goal will be to predict how many cylinders in a car

First we drop the character features:

$```\{$r$\}$

# Check if each column is character type

char$\_$columns $<-$ sapply$($Auto$,$ is$.$character$)$

# Print the results

print$($char$\_$columns$)$

$```$

$```\{$r$\}$

New$\_$Auto $<-$ select$\_$if$($Auto$,$ function$($x$)!$is$.$character$($x$))$

$```$

$```\{$r$\}$

summary$($New$\_$Auto$)$

$```$

Create training and Testing data: Complete the code

$```\{$r$\}$

#Create a $60/40$ training and testing split

intrain $<-$sample$(2$:nrow$($New$\_$Auto$),0\mathrm{.}6*$ nrow$($New$\_$Auto$))$

train$\_$data $<-$ New$\_$Auto$[$intrain $,]$

test$\_$data $<-$ New$\_$Auto$[-$intrain $,]$

$```$

$```\{$r$\}$

dim$($train$\_$data$)$

$```$

$```\{$r$\}$

summary$($train$\_$data$)$

$```$

$```\{$r$\}$

dim$($test$\_$data$)$

$```$

$```\{$r$\}$

summary$($test$\_$data$)$

$```$

Create a regression tree using only the training data:

$```\{$r$\}$

# Identify factor predictors with more than $32$ levels

factor$\_$predictors $<-$ sapply$($train$\_$data, is$.$factor$)$

levels$\_$count $<-$ sapply$($train$\_$data$[$factor$\_$predictors$],$ function$($x$)$ length$($levels$($x$)))$

high$\_$levels $<-$ names$($levels$\_$count$[$levels$\_$count $>32])$

# Convert factor predictors with more than $32$ levels to numeric

train$\_$data$\_$numeric $<-$ train$\_$data

train$\_$data$\_$numeric$[$high$\_$levels$]<-$ lapply$($train$\_$data$\_$numeric$[$high$\_$levels$],$ as$.$numeric$)$

#Create a tree using the $`$tree$()`$ function

TREE $<-$ tree$($mpg ~ $.,$ data $=$ train$\_$data$\_$numeric$)$

# Look at a summary of your tree

summary$($TREE$)$

$```$

How many nodes does it have?

# It has $8$ nodes.

Which variables did it find important?

# Weight, Horsepower, and Year.

Now plot your tree:

$```\{$r$\}$

plot$($TREE$)$

text$($TREE$,$ pretty $=0)$

$```$

Lets check it:Complete the code

$```\{$r$\}$

# Identify factor predictors with more than $32$ levels

factor$\_$predictors $<-$ sapply$($train$\_$data, is$.$factor$)$

levels$\_$count $<-$ sapply$($train$\_$data$[$factor$\_$predictors$],$ function$($x$)$ length$($levels$($x$)))$

high$\_$levels $<-$ names$($levels$\_$count$[$levels$\_$count $>32])$

# Convert factor predictors with more than $32$ levels to numeric

train$\_$data$\_$numeric $<-$ train$\_$data

train$\_$data$\_$numeric$[$high$\_$levels$]<-$ lapply$($train$\_$data$\_$numeric$[$high$\_$levels$],$ as$.$numeric$)$

# Remove 'name' variable from the dataset

train$\_$data$\_$numeric $<-$ train$\_$data$\_$numeric$[,!$names$($train$\_$data$\_$numeric$)\%$in$\%$ "name"$]$

# Create a tree using the $`$tree$()`$ function

TREE $<-$ tree$($mpg ~ $.,$ data $=$ train$\_$data$\_$numeric$)$

# Look at a summary of your tree

summary$($TREE$)$

$```$

$```\{$r$\}$

TREE$\_$hat $<-$ predict$($TREE$,$ newdata $=$ test$\_$data $)$

mean$(($TREE$\_$hat $-$ test$\_$data$mpg$)^2)$

$```$

$```\{$r$\}$

str$($train$\_$data$)$

$```$

Lets try random forest with m $=4$ and ntree $=5$: Complete the code $($remember we are predicting cylinders$)$

$```\{$r$\}$

# Convert categorical variables to factors

train$\_$data $<-$ lapply$($train$\_$data, function$($x$)\{$

if$($is$.$factor$($x$))$ x $<-$ as$.$factor$($x$)$

x

$\})$

# Perform one$-$hot encoding

train$\_$data$\_$encoded $<-$ model.matrix$($~ $.-1,$ data $=$ train$\_$data$)$

# Fit random forest model

ForestAuto $<-$ randomForest$($cylinders ~ $.,$ data $=$ train$\_$data, mtry $=4,$ importance $=$ TRUE, ntree $=5)$

# Print the random forest model

ForestAuto

$```$

Let's check it$!$Complete the code

$```\{$r$\}$

Forest$\_$hat $<-$ predict$()$

mean$(($Forest$\_$hat$-$test$\_$data$[,2])^2)$

$```$

Which one was better between a simple regression tree and random forest?