Memory efficient way to sample from a vector of counts without replacement

Question

Here I represent a jar of marbles using a vector of color frequencies

marbleCounts <- c(red = 5, green = 3, blue = 2)
marbleCounts

red green  blue 
  5     3     2 

Now, I'd like to sample 5 marbles from this vector without replacement. I can do this by expanding my vector of frequencies into a vector of marbles and then sampling from it.

set.seed(2019)
marbles <- rep(names(marbleCounts), times = marbleCounts)
samples <- sample(x = marbles, size = 5, replace = FALSE)
table(samples)

green   red 
    2     3 

but this is memory inefficient (and perhaps performance inefficient?). Is there a faster and/or more efficient way to sample data like this?

Answer 1

I think this will work for you.

marbleCounts <- c(red = 5, green = 3, blue = 2)

# first, draw from the possible indexes (does not create the full vector)
draw <- sample.int(sum(marbleCounts), 5)

# then assign indexes back to original group
items <- findInterval(draw-1, c(0, cumsum(marbleCounts)), rightmost.closed = TRUE)

#extract your sample    
obs <- names(marbleCounts)[items]
table(obs)

This will never create a vector longer than your sample size.