T-SVM.R

sample.data <- read.csv("svm_sample.csv")
sample.data <- sample.data[,-1] #getting rid of id variables


library(caret)
train_index <- createDataPartition(sample.data$color, 0.8)[[1]]

sample.data.train <- sample.data[train_index,]
sample.data.test <- sample.data[-train_index,]

##########################################################################
## MODEL 1:
##doing a linear svm
##########################################################################

sample.svm <- svm(color ~ ., data = sample.data.train, kernel = "linear")

sample.svm #creates 75 support vectors!!!
svm.predictions_linear <- predict(sample.svm, sample.data.test)

table(svm.predictions_linear,sample.data.test$color)


##########################################################################
## MODEL 2:
##doing a radial svm
##########################################################################


##doing a radial svm
library(e1071)
sample.svm <- svm(color ~ ., data = sample.data.train, kernel = "radial")

sample.svm #creates 75 support vectors!!!
svm.predictions_radial <- predict(sample.svm, sample.data.test)

table(svm.predictions_radial,sample.data.test$color)

###TUNING THIS MORE

?tune

svm_tune <- tune.svm(color ~., data = sample.data.train, kernel="radial", cost=10^(-1:2), gamma=c(.5,1,2))

print(svm_tune)

tuned_svm <- svm(color ~., data = sample.data.train, kernel = "radial", cost = 0.1, gamma = 2)

svm.predictions_tuned_radial <- predict(tuned_svm, sample.data.test)

table(svm.predictions_tuned_radial, sample.data.test$color)


##########################################################################
## MODEL 3:
##doing a logistic regression
##########################################################################

#need to convert values to a 0 or 1 for a logistic regression
sample.data.train.log <- sample.data.train
sample.data.test.log <- sample.data.test

sample.data.train.log$color <- as.numeric(sample.data.test$color) - 1
sample.data.test.log$color <- as.numeric(sample.data.test$color) - 1
## this means purple is 0, green is 1

sample.logistic <- glm(color ~ ., data = sample.data.train.log, family = "binomial")
log.predictions <- predict(sample.logistic, sample.data.test.log, link = "response")
log.predictions <- ifelse(log.predictions<0.5, 0, 1)
log.predictions[log.predictions==0] <- "purple"
log.predictions[log.predictions==1] <- "green"

table(log.predictions, sample.data.test.log$color) #utter disaster



##########################################################################
## MODEL 4:
##doing a logistic regression
##########################################################################
##doing a decision tree

sample.rpart <- rpart(color ~ ., data = sample.data.train)
library(rattle)
fancyRpartPlot(sample.rpart)

sample.rpart #creates 75 support vectors!!!
rpart.predictions <- predict(sample.rpart, sample.data.test, type = "class")


## assessing accuracy for all models
accuracy_rpart <- sum(rpart.predictions==sample.data.test$color)/length(sample.data.test$color)
accuracy_svmlinear <- sum(svm.predictions_linear==sample.data.test$color)/length(sample.data.test$color)
accueracy_svmradial_tuned <- sum(svm.predictions_tuned_radial==sample.data.test$color)/length(sample.data.test$color)
accuracy_svmradial <- sum(svm.predictions_radial==sample.data.test$color)/length(sample.data.test$color)
accuracy_log <- sum(log.predictions==sample.data.test.log$color)/length(sample.data.test$color)