I realize now that I may have been in error - at least RandomEngine and MersenneTwister have been undeprecated (although I can't see where they are used aside from tests), so maybe they're still usable.
Ted, have you looked over the various random stuff (hah!) we have in org.apache.mahout.math.jet.random.* ? On Tue, Apr 26, 2011 at 1:29 PM, Randall McRee <[email protected]>wrote: > Sid might want to make it pluggable, thats fine but that will require > settling on an interface we believe in. Since RandomEngine is deprecated > should that be org.apache.commons.math. > random.RandomGenerator? > > Going forward folks like me are going to want to add *many* more classes > that depend on whatever this parent interface is. Hence, I would like to > contribute something that is forward-looking. > > If you look at MonteCarlo open source and genetic programming frameworks > practically every interesting feature will depend directly on this parent > interface. > > Just for reference, RandomGenerator has a nice minimum set of abstract > methods: setSeed(), nextDouble(), nextGaussian(), nextFloat(), nextLong(), > nextInt(), nextInt(int) > > I would like to write to this set, or something close to it. > > Randy > > > On Tue, Apr 26, 2011 at 12:22 PM, Jake Mannix <[email protected]> > wrote: > > > On Tue, Apr 26, 2011 at 12:19 PM, Sid <[email protected]> wrote: > > > > > May be make it implementation specific... For Unix/Linux systems > atleast > > > the > > > implementation should perhaps poll /dev/random which is truly random > and > > > for > > > windows settle for the best one? > > > > > > > /dev/random is way too slow for production use, as it blocks on the > entropy > > pool > > available. For statistical purposes, /dev/urandom is totally fine. > > > > -jake > > >
