Sorting of strata in training data from initial_split

[dicook]

I have just completed running a kaggle challenge for my machine learning class. I had a surprise that the training data is sorted by the variable used for strata:

# Here is code to reproduce
set.seed(921)
d <- tibble(x=runif(100), y=sample(c("y", "n"), 100, replace=TRUE))
d_split <- initial_split(d, strata=y)
d_tr <- training(d_split)
d_tr$y
d_ts <- testing(d_split)
d_ts$y

and we find

> d_tr$y
 [1] "n" "n" "n" "n" "n" "n" "n" "n" "n" "n" "n" "n"
[13] "n" "n" "n" "n" "n" "n" "n" "n" "n" "n" "n" "n"
[25] "n" "n" "n" "n" "n" "n" "n" "n" "n" "n" "n" "n"
[37] "n" "n" "y" "y" "y" "y" "y" "y" "y" "y" "y" "y"
[49] "y" "y" "y" "y" "y" "y" "y" "y" "y" "y" "y" "y"
[61] "y" "y" "y" "y" "y" "y" "y" "y" "y" "y" "y" "y"
[73] "y" "y"

although test set is not (luckily for me!)

d_ts$y
 [1] "n" "n" "n" "n" "y" "y" "n" "y" "y" "y" "y" "y"
[13] "y" "y" "y" "n" "n" "n" "n" "n" "y" "y" "n" "y"
[25] "n" "n"

I'm not sure if this is intentional but it would be better to have the default be that these are in a random order, with an input parameter to sort being optional.