install.packages('devtools')

library(devtools)
install_github("jlaria/sglfast")
library(sglfast)

library(sglfast)

# We create beta="the true coefficient vector" to be used in the simulations.
beta = 1:5

# We generate the model matrix X with iid columns and rows and the response y
X = matrix(rnorm(100*400), nrow = 100)
y = X[,1:5]%*%beta

# We chose the variance of the error such that SNR = 3
snr = 3
error = rnorm(100, mean = 0, sd=sqrt(var(y)/snr))
y = y+error

# Rows in the training sample
train.idx = sample(100, 50)

# Group indices for the SGL  
group_index = rep(1:40, each=10)

# Input data for the iterative 
data.train = list(x=X[train.idx,], y=y[train.idx])
data.validate = list(x=X[-train.idx,], y=y[-train.idx])

# We run the (unpooled) iterative SGL. For the 2-parameter version use isg_simple()
isgl.fit = isgl(data.train, data.validate, group_index, type = "linear")

# Best model returned by the iSGL algorithm
isgl.fit$beta
isgl.fit$intercept

library(sglfast)

# We create beta="the true coefficient vector" to be used in the simulations.
beta = 1:5

# We generate the model matrix X with iid columns and rows and the response y
X = matrix(rnorm(100*400), nrow = 100)
y = X[,1:5]%*%beta

# We generate the response from a logit model

y = ((1+exp(-y))^-1 > 0.5) + 0


# Rows in the training sample
train.idx = sample(100, 50)

# Group indices for the SGL  
group_index = rep(1:40, each=10)

# Input data for the iterative 
data.train = list(x=X[train.idx,], y=y[train.idx])
data.validate = list(x=X[-train.idx,], y=y[-train.idx])

# We run the (unpooled) iterative SGL. For the 2-parameter version use isg_simple()
isgl.fit = isgl_simple(data.train, data.validate, group_index, type = "logit")

# Best model returned by the iSGL algorithm
isgl.fit$beta
isgl.fit$intercept