# package for big n methods
library(datadr)
# divide the data into separate chunks to be processed
rrSA<- divide(SAheart, by = rrDiv(100), update = TRUE)

# BLB method
###### Here you choose your own statistics of interest
###### I am asking for the L and U limits of a 95% CI for each regression coefficient here
###### but you could ask for the estimates also
SABlb <- rrSA %>% addTransform(function(x) {
  drBLB(x,
statistic = function(x, weights)
       coef(lm(ldl ~ .,
         data = x, weights = weights)),
     metric = function(x)
       quantile(x, c(0.05, 0.95)),
     R = 100,
     n = nrow(rrSA)
   )
 })
######
coefs <- recombine(SABlb, combMean)

# compare BLB and full data results
matrix(coefs, ncol = 2, byrow = TRUE)
       
confint(lm(ldl~.,data=SAheart))

######################################################################3

### Leveraging
library(rsvd)
# random SVD for fast computation
data(spam)
ss<-rsvd(as.matrix(spam[,-58]))
lev<-apply(ss$u^2,1,sum)
# the leverage values
# sampling 500 observations with different probabilities given by leverage
probs<-lev*500/sum(lev)
probs[probs>1]<-1
pp<-rbinom(4601,prob=probs,size=1)

# Leverage results
mm<-glm(as.factor(chd)~., data=SAheart,subset=seq(1,4601)[pp==1],family=binomial)
coef(mm)