On Jul 17, 2005, at 1:11 PM, Joost Diepenmaat wrote:

> b) As soon as samples.size > total.size / 2 it's trivial to turn the
> algorithm on its head (i.e. randomly select items that you'll skip,
> then loop through the set and print all items that aren't selected).

That's a massive slowdown though, right?  Because you now need to  
count to 1_000_000_000 (for the quiz example.)

I agree that switching strategies is a good idea, but there are  
better strategies for the high members counts and/or memory  
problems.  Take a look through the submissions...

James Edward Gray II