$s$ balls are sampled and then replaced from/into a box with $N$ balls. On average, how many samplings are to be made until all of the balls have been taken at least once?
ps: this is not a homework problem. I’m trying to reduce the number of updates on parameters during a neural network’s optimization phase. I’m able to simulate this experiment and get a good numerical approximation, but doing so for every different value of $s$ and $N$ seems a little unclever.