Random Forest.Rmd

---
title: "Random Forest"
author: "Devan Goto"
date: "October 9, 2017"
output: html_document
---

```{r}
# 0. Import the data into the notebook/environment. 

a1<-read.csv("agaricus-lepiota.data.txt",header=FALSE,sep=",")

View(a1)

a2<-a1

```

```{r}
# 1. Read the file and check for any null values

length(which(a2$stalk.root=="?"))

```

```{r}
#2. Add column names as specified in  https://archive.ics.uci.edu/ml/machine-learning-databases/mushroom/

colnames(a2) [1]<-"edible.poisen"
colnames(a2) [2]<-"cap.shape"
colnames(a2) [3]<-"cap.surface"
colnames(a2) [4]<-"cap.color"
colnames(a2) [5]<-"bruises"
colnames(a2) [6]<-"odor"
colnames(a2) [7]<-"gill.attachment"
colnames(a2) [8]<-"gill.spacing"
colnames(a2) [9]<-"gill.size"
colnames(a2) [10]<-"gill.color"
colnames(a2) [11]<-"stalk.shape"
colnames(a2) [12]<-"stalk.root"
colnames(a2) [13]<-"stalk.surface.above.ring"
colnames(a2) [14]<-"stalk.surface.below.ring"
colnames(a2) [15]<-"stalk.color.above.ring"
colnames(a2) [16]<-"stalk.color.below.ring"
colnames(a2) [17]<-"veil.type"
colnames(a2) [18]<-"veil.color"
colnames(a2) [19]<-"ring.number"
colnames(a2) [20]<-"ring.type"
colnames(a2) [21]<-"spore.print.color"
colnames(a2) [22]<-"population"
colnames(a2) [23]<-"habitat"

```

```{r}
#3. Print the structure of the data

print(str(a2))

```

```{r}
#4. Print the head of the data set

print(head(a2))

```

```{r}
#5. Using a core tidyverse package, try to fit the tibble in the screen

install.packages("tidyverse")
library(tidyverse)
as_tibble(a2)

```

```{r}
#6. Using a core tidyverse package, try to add row names as well

a3<-a2

#Creates a new column that represents each instance/participant/mushroom/etc.
a3$rowname<-1:nrow(a3)

#Moves rownames variable to the front.
a3<-a3[c(24,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23)]

```

```{r}
#7. Print the all columns for which class is edible

edible1<-subset(a2, a2$`edible-poisen` == "e")

```

```{r}
#8 Split the data into training and test sets of 80 -20 distribution

#training and testing for whole dataset
training<-dplyr::sample_frac(a3, .8)
testing<-dplyr::setdiff(a3, training)
  
#training and testing for only edible items
edible.training<-dplyr::sample_frac(edible1,.8)
edible.testing<-dplyr::setdiff(edible1, edible.training)

```

```{r}
#9. Use random forest on the train set and predict the top 10 variables of importance. Try the same on test set. Do the results vary?

#What:
##An ensemble learning method for classification and regression
##Operate by constructing a multitude of decision trees 

#Why use Random Forest:
##Reasonable fast but very easy to use
##Handles sparse data/missing data well
##Overcomes problem with over fitting

#How:
##Tree bagging - random sample with replacement 
##Random subset of the features.
##Voting: Algorithm creates mutiple decision trees, and votes to choose outcome based on multiple trees.

install.packages("randomForest")
library("randomForest")

#for training
rf.edible<-randomForest(edible.poisen ~., training)

#see overview info on randomForest
print(rf.edible)

#the higher the value the more important. useful for feature selection.
importance(rf.edible)

#see tree
getTree(rf.edible,labelVar =TRUE)

#for testing
rf.edible2<-randomForest(edible.poisen ~., testing)
print(rf.edible2)
importance(rf.edible2)
getTree(rf.edible2,labelVar =TRUE)

```

```{r}
#10. Print the top 10 variables columns for which class is edible


#need to install a newer version of R to install this package
install.packages("varimplot")
library("varimplot")

#Shows top 10 Variables in Training & Testing
varImplPlot(rf.edible,sort=T,main="Top 10 Variables Training", n.var=10)
varImplPlot(rf.edible2,sort=T,main="Top 10 Variables Testing", n.var=10)

#Shows how each variable is predicting edible. 
v.train.importance<-data.frame(importance(rf.edible))
v.train.importance2<-data.frame(importance(rf.edible2,type = 2))

```