Basics of R- Session 15 Multiple Regression 2

# Machine Learning Regression Ridge lasso

install.packages(c("caret", ";", "glmnet;", "mlbench;", "psych"))

library(caret)
library(glmnet)
library(mlbench)
library(psych)
library(ggplot2)

# data set, reg2

reg1<-read.csv("C:/Users/Administrator.vaio1/Desktop/reg2.csv")
# keep only those variables which are suppose to be used in analysis

reg1X<-reg1[,8:19]
reg1X<-as.matrix(reg1X)
fix(reg1X)

reg1Y<-as.matrix(reg1[,20])
fix(reg1Y)

lambda1<-10^seq(10,-2,length =100)

ridge1<-glmnet(reg1X, reg1Y, alpha = 0, lambda = lambda1)
plot(ridge1)

ridge1$lambda[1]
coef(ridge1)[,1] # dislay the first model

ridge1$lambda[2]
coef(ridge1)[,2] # dislay the first model

ridge1$lambda[100]
coef(ridge1)[,100] # dislay the first model




lasso1<-glmnet(reg1X, reg1Y, alpha = 1, lambda = lambda1)
plot(ridge1)

lasso1$lambda[1]
coef(lasso1)[,1] # dislay the first model

lasso1$lambda[2]
coef(lasso1)[,2] # dislay the first model

lasso1$lambda[100]
coef(lasso1)[,100] # dislay the first model

#-------------------------------------#

#--------------------------------------------------------------#

# Machine Learning Regression Ridge lasso

install.packages(c("caret", ";", "glmnet;", "mlbench;", "psych"))

library(caret)

library(glmnet)

library(mlbench)

library(psych)

library(ggplot2)

reg1<-read.csv("C:/Users/Administrator.vaio1/Desktop/reg2.csv")

reg1<-reg1[,8:20]

# linear regression

lm<-train(X19~.,reg1, method= "lm")

summary(lm)

# ridge regression

ridge1<-train(X19~.,reg1, method= "glmnet", 

              tuneGrid=expand.grid(alpha =0, lambda = seq(0.0001,1,length=100)))

ridge1

plot(ridge1)

plot(ridge1$finalModel)

plot(ridge1$finalModel, xvar = "lambda", label = TRUE)

# parameters alpha and lambda

ridge

ridge$bestTune

# coeficient of best model

coef(ridge$finalModel)

coef(ridge$finalModel, s=ridge$bestTune$lambda) # for which value of lambda

coef(ridge$finalModel, s=ridge$bestTune$alpha) # for which value of alpha

plot(varImp(ridge1))

# best model of ridge

ridge1$bestTune

bestridge<-ridge1$finalModel

coef(bestridge, s=ridge1$bestTune$lambda)

# lasso regression

lasso1<-train(X19~.,reg1, method= "glmnet", 

              tuneGrid=expand.grid(alpha =1, lambda = seq(0.0001,1,length=100)))

lasso1

plot(lasso1)

plot(lasso1$finalModel)

plot(lasso1$finalModel, xvar = "lambda", label = TRUE)

plot(lasso1$finalModel, xvar = "dev", label = TRUE)

plot(varImp(lasso1))

# best model of lasso

lasso1$bestTune

bestlasso<-lasso1$finalModel

coef(bestlaso, s=lasso1$bestTune$lambda)

# elastic net regression

en1<-train(X19~.,reg1, method= "glmnet", 

              tuneGrid=expand.grid(alpha =seq(0,1, length=10), lambda = seq(0.0001,1,length=100)))

plot(en1)

plot(en1$finalModel)

plot(en1$finalModel, xvar = "lambda", label = TRUE)

plot(en1$finalModel, xvar = "dev", label = TRUE)

plot(varImp(en1))

# best model of elastic net

en1$bestTune

besten<-en1$finalModel

coef(besten, s=en1$bestTune$lambda)