将数据拆分为在标签上分层的训练和测试

Question

I have a data frame (df) with two columns (Numbers and Letters). See reproducible example:

Numbers<- c(2.370653,3.811336,5.255120, 6.501197,7.848100,9.343938,10.843479,12.164387,13.476807,14.922644,16.419281,17.664224,19.112835,20.660367,21.962732,23.213675)
Letters<-c("a","b","c","c","d","a","b","d","d","a","a","c","b","c","c","c")
df <- as.data.frame(cbind(Numbers,Letters))

I want randomly to split the data frame into two date frames of equal size and with the same number of Letters in each. I have found the stratified() function that takes a sample with 50% of each of the Letters:

test <- stratified(df, "Letters", .5)

But this is not really the same as splitting the data frame into two data frames. I do not want any of the same values from df$Numbers in the two data frames - just the same amount of df$Letters in each. Can you help me?

Answer 1

Try this approach with rsample which is close to what you want. And the comment of @AllanCameron is totally valid, you can split three into two pieces of 1.5 for each sample:

library(rsample)
#Code
set.seed(123)
split_strat <- initial_split(df, prop = 0.5,
                             strata = 'Letters')
train_strat <- training(split_strat)
test_strat <- testing(split_strat)

Check for proportions:

table(train_strat$Letters)

a b c d 
2 2 3 2 

table(test_strat$Letters)

a b c d 
2 1 3 1