I'm using latin hypercube sampling to draw parameter values for an epidemiological simulation model (about 30 parameters, following various distributions), and so far I haven't been able to find any good rule of thumb/heuristics/theory for choosing the number of samples (i.e. the number of equiprobable intervals).
Asked
Active
Viewed 555 times
6
-
4Your question doesn't contain a question. – naught101 Mar 08 '12 at 01:30
-
Although there is an implied question, there does not seem to be enough information to answer it. – whuber May 22 '16 at 20:38
-
I think the problem is a bit unclear, but at least we can say that LHS puts no real constraints on the number of points N. But if you have that many parameters and if no single parameter dominates, then LHS will give not that much variance reduction anyway. Best make some tests runs (e.g. N=50 - often enough if you know quite well what you can expect), then decide if you need more points or not. – user32038 Aug 25 '17 at 08:25