random subset of fixed length such that each group is present at least N times

Question

I want to select 5 rows from each values in column1 of df such that the output has at least 1 value for each unique value in column2. Also there should not be any duplicates in the output

Edit: There should not be duplicates in (column1, column3) pair: i.e For each value in column1 all values in column3 should be unique

column1 = rep(c("a","b"), each = 12)
column2 = rep(c(1,2,3), each = 4)
column3 = c("x1","x2","x3","x4","x5","x3","x6","x7","x8","x1","x9","x5","x6","x2","x3","x4","x7","x5","x6","x1","x4","x1","x6","x9")

df = data.frame(column1, column2, column3)

Here is a valid solution

sample_output_1 = data.frame(column1 = rep(c("a","b"), each = 5),
                         column2 = c(1,1,2,2,3,1,1,2,2,3),
                         column3 = c("x1","x2","x5","x3","x8","x6","x2","x5","x1","x9"))

Answer 1

Check this

foo = function(a_df){
    inds = 1:NROW(a_df)     
    #Sample 5 indices along the rows of a_df
    my_inds = sample(inds, 5)       
    #If subset of a_df based on my_inds has duplicates
    #Or if 2nd column does not have all unique values
    while(any(duplicated(a_df[my_inds, c(1, 3)])) & 
        !identical(sort(unique(a_df[my_inds, 2])), sort(unique(a_df[[2]])))){
            #Count the number of duplicates or missing all values
            n = sum(duplicated(a_df[my_inds, c(1, 3)]))
            n = n + sum(!sort(unique(a_df[my_inds, 2])) %in% sort(unique(a_df[[2]])))
            #Remove my_inds from inds
            inds = inds[!inds %in% my_inds]
            #Remove the n indices that create duplicates from my_nds
            my_inds = my_inds[!duplicated(a_df[my_inds, c(1, 3)])]
            #Sample n more from inds and add to my_inds
            my_inds = sample(c(my_inds, sample(inds, n)))
        }
        return(a_df[my_inds,])
}

set.seed(42)
do.call(rbind, lapply(split(df, df$column1), function(a) foo(a_df = a)))
     # column1 column2 column3
# a.11       a       3      x9
# a.12       a       3      x5
# a.3        a       1      x3
# a.8        a       2      x7
# a.6        a       2      x3
# b.19       b       2      x6
# b.21       b       3      x4
# b.14       b       1      x2
# b.18       b       2      x5
# b.23       b       3      x6