# install the package, then load it
library(nn4pse)

?dat # see Keller, Kim, & Steiner (2015) for more details
head(dat)
hist(dat$ps) # true propensity scores
hist(dat$ps[dat$sel == 0]) # PSs for untreated
hist(dat$ps[dat$sel == 1]) # PSs for treated

# Define ps formula:
psform <- as.formula(paste0("sel.fact ~ ", paste0("X", 1:12, collapse = " + ")))
psform

# Set number of cores to number available on your machine
doMC::registerDoMC(cores = 4)

# Fit the nn model
out1 <- cv_train(ps_formula = psform, data = dat, folds = 5, decay = 10^seq(-4, 0, 1), size = c(2, 4, 8, 16), ncores = 4)
out1$Performance
out1$BestTune

# Histogram of estimated PSs (compare with actual PSs above)
hist(out1$PropensityScore)
hist(out1$PropensityScore[dat$sel == 0])
hist(out1$PropensityScore[dat$sel == 1])
cor(dat$ps, out1$PropensityScore)