## k-NN for IRIS dataset. The file content is included at the bottom of this markdown
source('knn_functions.R')
str(iris)
## 'data.frame': 150 obs. of 5 variables:
## $ Sepal.Length: num 5.1 4.9 4.7 4.6 5 5.4 4.6 5 4.4 4.9 ...
## $ Sepal.Width : num 3.5 3 3.2 3.1 3.6 3.9 3.4 3.4 2.9 3.1 ...
## $ Petal.Length: num 1.4 1.4 1.3 1.5 1.4 1.7 1.4 1.5 1.4 1.5 ...
## $ Petal.Width : num 0.2 0.2 0.2 0.2 0.2 0.4 0.3 0.2 0.2 0.1 ...
## $ Species : Factor w/ 3 levels "setosa","versicolor",..: 1 1 1 1 1 1 1 1 1 1 ...
summary(iris)
## Sepal.Length Sepal.Width Petal.Length Petal.Width
## Min. :4.300 Min. :2.000 Min. :1.000 Min. :0.100
## 1st Qu.:5.100 1st Qu.:2.800 1st Qu.:1.600 1st Qu.:0.300
## Median :5.800 Median :3.000 Median :4.350 Median :1.300
## Mean :5.843 Mean :3.057 Mean :3.758 Mean :1.199
## 3rd Qu.:6.400 3rd Qu.:3.300 3rd Qu.:5.100 3rd Qu.:1.800
## Max. :7.900 Max. :4.400 Max. :6.900 Max. :2.500
## Species
## setosa :50
## versicolor:50
## virginica :50
##
##
##
set.seed(100)
m <- avgTrnTst(iris, 0.7, 5)
dim(m)
## [1] 40 3
plotFn(m, 'Training and Testing Accuracy for k-NN of Iris data-set')
## Decision Tree for Iris dataset
library(rpart)
library(rpart.plot)
v <- iris$Species
table(v)
## v
## setosa versicolor virginica
## 50 50 50
set.seed(522)
# runif function returns a uniform distribution which can be further conditionally split into 75-25 ratio
iris[, 'train'] <- ifelse(runif(nrow(iris)) < 0.75, 1, 0)
trainSet <- iris[iris$train == 1,]
testSet <- iris[iris$train == 0, ]
trainColNum <- grep('train', names(trainSet))
trainSet <- trainSet[, -trainColNum]
testSet <- testSet[, -trainColNum]
treeFit <- rpart(Species~.,data=trainSet,method = 'class')
print(treeFit)
## n= 111
##
## node), split, n, loss, yval, (yprob)
## * denotes terminal node
##
## 1) root 111 74 setosa (0.33333333 0.33333333 0.33333333)
## 2) Petal.Length< 2.45 37 0 setosa (1.00000000 0.00000000 0.00000000) *
## 3) Petal.Length>=2.45 74 37 versicolor (0.00000000 0.50000000 0.50000000)
## 6) Petal.Width< 1.75 39 2 versicolor (0.00000000 0.94871795 0.05128205) *
## 7) Petal.Width>=1.75 35 0 virginica (0.00000000 0.00000000 1.00000000) *
rpart.plot(treeFit, box.col=c("red", "green"))
Prediction1 <- predict(treeFit,newdata=testSet[-5],type = 'class')
## Print the confusion matrix to check the accuracy and other statistics
library(caret)
## Loading required package: lattice
## Loading required package: ggplot2
confusionMatrix(Prediction1,testSet$Species)
## Confusion Matrix and Statistics
##
## Reference
## Prediction setosa versicolor virginica
## setosa 13 0 0
## versicolor 0 12 3
## virginica 0 1 10
##
## Overall Statistics
##
## Accuracy : 0.8974
## 95% CI : (0.7578, 0.9713)
## No Information Rate : 0.3333
## P-Value [Acc > NIR] : 3.435e-13
##
## Kappa : 0.8462
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: setosa Class: versicolor Class: virginica
## Sensitivity 1.0000 0.9231 0.7692
## Specificity 1.0000 0.8846 0.9615
## Pos Pred Value 1.0000 0.8000 0.9091
## Neg Pred Value 1.0000 0.9583 0.8929
## Prevalence 0.3333 0.3333 0.3333
## Detection Rate 0.3333 0.3077 0.2564
## Detection Prevalence 0.3333 0.3846 0.2821
## Balanced Accuracy 1.0000 0.9038 0.8654
## Pruning the decision tree
printcp(treeFit)
##
## Classification tree:
## rpart(formula = Species ~ ., data = trainSet, method = "class")
##
## Variables actually used in tree construction:
## [1] Petal.Length Petal.Width
##
## Root node error: 74/111 = 0.66667
##
## n= 111
##
## CP nsplit rel error xerror xstd
## 1 0.50000 0 1.000000 1.148649 0.060298
## 2 0.47297 1 0.500000 0.783784 0.071115
## 3 0.01000 2 0.027027 0.027027 0.018938
opt <- which.min(treeFit$cptable[,'xerror'])
cp <- treeFit$cptable[opt, 'CP']
pruned_model <- prune(treeFit,cp)
rpart.plot(pruned_model, box.col=c("red", "green"))
rpart_pruned_predict <- predict(pruned_model, newdata=testSet[-5],type = 'class')
mn2 <- mean(rpart_pruned_predict==testSet$Species)
mn2
## [1] 0.8974359
confusionMatrix(rpart_pruned_predict,testSet$Species)
## Confusion Matrix and Statistics
##
## Reference
## Prediction setosa versicolor virginica
## setosa 13 0 0
## versicolor 0 12 3
## virginica 0 1 10
##
## Overall Statistics
##
## Accuracy : 0.8974
## 95% CI : (0.7578, 0.9713)
## No Information Rate : 0.3333
## P-Value [Acc > NIR] : 3.435e-13
##
## Kappa : 0.8462
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: setosa Class: versicolor Class: virginica
## Sensitivity 1.0000 0.9231 0.7692
## Specificity 1.0000 0.8846 0.9615
## Pos Pred Value 1.0000 0.8000 0.9091
## Neg Pred Value 1.0000 0.9583 0.8929
## Prevalence 0.3333 0.3333 0.3333
## Detection Rate 0.3333 0.3077 0.2564
## Detection Prevalence 0.3333 0.3846 0.2821
## Balanced Accuracy 1.0000 0.9038 0.8654
knn_functions.R file content
# 'caTools' package provides us with functions to split dataset uniformly to test and training
library(caTools)
# Load library 'class' that has the knn() function
library(class)
# Function to split the dataset randomly
splitFile <- function(dataset, trProp, classColPos) {
# split the dataset
sample = sample.split(iris[, classColPos], SplitRatio = trProp)
# create training and testing dataset
train = subset(iris, sample == TRUE)
test = subset(iris, sample == FALSE)
# save the target labels and remove from the train and test dataset
trainLabels <- train[, classColPos]
testLabels <- test[, classColPos]
train <- train[, -classColPos]
test <- test[, -classColPos]
# Nomalize function
normalize <- function(x) {
return( (x-min(x))/(max(x)-min(x)))
}
train
test
# Normalize test and training dataset
gtrn <- as.data.frame(lapply(train, normalize))
gtsn <- as.data.frame(lapply(test, normalize))
return(list(trn=gtrn, trL=trainLabels, val=gtsn, tsL=testLabels))
}
# Function to plot graph
plotFn <- function(dataSet, graphTitle = '', ylimLo=0) {
plot(dataSet[, 1], dataSet[, 2], main = graphTitle, xlab = 'k Nearest Neighbours',
ylab = 'Accuracy', ylim = c(ylimLo, 1), type = 'o', col = 'red')
lines(dataSet[, 1], dataSet[, 3], type = 'o', col = 'blue')
legend(26, 0.6, legend=c("Training Accuracy", "Testing Accuracy"),
col=c("red", "blue"), lty=1:2, cex=1.4)
}
# Function to use k-NN and return training and testing results
train_test <- function(trainData,trainLabels,testData,testLabels) {
train <- c()
test <- c()
for (k in 1:40) {
knntr <- knn(trainData, trainData, trainLabels, k=k)
knnts <- knn(trainData, testData, trainLabels, k=k)
trTable <- table(knntr, trainLabels)
tsTable <- table(knnts, testLabels)
trTable <- prop.table(trTable)
tsTable <- prop.table(tsTable)
trainAccuracy <- sum(trTable[1,1], trTable[2,2], trTable[3,3])/sum(trTable)
testAccuracy <- sum(tsTable[1,1], tsTable[2,2], trTable[3,3])/sum(tsTable)
train <- c(train, trainAccuracy)
test <- c(test, testAccuracy)
}
acc <- data.frame('k' = 1:40, 'trAc' = train, 'tsAc' = test)
return(acc = acc)
}
# Single function to split data and then call train_test function
avgTrnTst <- function(dataset, trProp, classColPos) {
for (i in 1:30) {
a <- splitFile(dataset, trProp, classColPos)
b <- train_test(a$trn, a$trL, a$val, a$tsL)
if (i==1) acd <- b
else acd <- rbind(acd, b)
}
library(plyr)
a1 <- ddply(acd,.(k), summarize, meanV = mean(trAc))
a2 <- ddply(acd,.(k), summarize, meanV = mean(tsAc))
m <- merge(a1,a2,by='k')
return(m)
}