This function is used to perform stratified random sampling to balance outcomes among the shards.
stratrs(y, C=5, P=0)A vector is returned with each element assigned to a shard.
The binary/categorical/continuous outcome.
The number of shards to break the data set into.
For continuous data, we break the range into P segments via the quantiles. Specifying, P=20 seems to work reasonably well.
To perform BART with large data sets, random sampling is employed
to break the data into C shards. Each shard should be
balanced with respect to the outcome. For binary/categorical
outcomes, stratified random sampling is employed with this function.
rs.pbart
set.seed(12)
x <- rbinom(25000, 1, 0.1)
a <- stratrs(x)
table(a, x)
z <- pmin(rpois(25000, 0.8), 5)
b <- stratrs(z)
table(b, z)
Run the code above in your browser using DataLab