Han de Bruijn wrote: > regis lebrun wrote: > > > [ ... ] but I can't use it if I want to generate a really huge > > permutation (let's say n=10^9). > > Instead of asking more than seven wise men can answer, wouldn't you try > to consider the possibility of sizing down your problem a little bit .. > > Han de Bruijn
First of all, thanks for this quick answer.
Of course I could tell the users to perform their Monte Carlo-like simulations with less samples, but I don't really think it is the right answer ;-). You know, if someone is able to compute some probability with pure Monte Carlo by throwing 10^9 times the dices and evaluating a cheap analytical function over this huge sample (it takes less than 1mn in Matlab, standard PC computer), he will try to do the same with LHS, and then, BOOM->out of memory.
By the way, I imagine that the underlying algorithmic problem is interresting: how much information (i.e. how much memory) do I need to store to be able to write down a permutation of [1, ..., n] without storing it explicitely?