## R commands for STAT 485
## Intermediate Statistical Techniques for Machine Learning and Big Data
## Version: 2024.01.26
## Reference: http://homepages.math.uic.edu/~jyang06/stat485/stat485.html


## R package: ElemStatLearn, "ElemStatLearn_2015.6.26.2.tar"
## Source: https://cran.r-project.org/src/contrib/Archive/ElemStatLearn/


## Data Sets, Functions and Examples from the Book (ESL): 
## "The Elements of Statistical Learning, Data Mining, Inference, and Prediction" 
## by Trevor Hastie, Robert Tibshirani and Jerome Friedman, Second Edition, 2017 (corrected 12th printing)
## Source: https://hastie.su.domains/ElemStatLearn/download.html

## Reference: Section 3.8.4 in the ESL book

## install the package  "grpreg"
# install.packages("grpreg")

## Birth weight data: Raw data in "MASS"
library(MASS)
data(birthwt)
dim(birthwt)  # 189  10
str(birthwt)

## R package: grpreg
library("grpreg")

## Birth weight data: Raw data in "grpreg"
data(Birthwt)
X <- Birthwt$X         # Matrix of predictors
dim(X)  # 189  16

group <- Birthwt$group # Vector describing how the columns of X are grouped

## Group lasso: Linear regression
y <- Birthwt$bwt
fit <- grpreg(X, y, group, penalty="grLasso")
plot(fit, label=T)   # plot coefficients with predictor labels
plot(fit, norm=TRUE, label=TRUE)   # plot coefficients with group labels

# choose tuning parameter
set.seed(325)
cvfit <- cv.grpreg(X, y, group, nfolds=10, penalty="grLasso")
plot(cvfit, norm=TRUE, label=TRUE)
summary(cvfit)
coef(cvfit) ## Beta at minimum CVE

## Group lasso: Logistic regression
y.low <- Birthwt$low
fit.logit <- grpreg(X, y.low, group, penalty="grLasso", family="binomial")
plot(fit.logit, norm=TRUE, label=TRUE)

# choose tuning parameter
set.seed(598)
cvfit.logit <- cv.grpreg(X, y.low, group, nfolds=10, penalty="grLasso", family="binomial")
plot(cvfit.logit, norm=TRUE, label=TRUE)
summary(cvfit)
coef(cvfit) ## Beta at minimum CVE