On 7/19/12 7:16 PM, Ted Dunning wrote: > One thing that should be considered is how a deterministic seed can be > injected during testing to make tests deterministic.
That is one reason to keep the reseed method that exists now. That method provides a seed to the underlying PRNG. The "default" Gilles chose above (which I think is fine - sorry for non-response) generates a fixed sequence when started with a fixed seed. Phil > > Mahout does this by providing a static method that returns a standard PRNG. > During tests, that PRNG is seeded by a constant. During normal operation, > the PRNG is seeded using system entropy from /dev/urandom (not /dev/random > since we don't want to block waiting for initialization) > > On Thu, Jul 19, 2012 at 5:43 PM, Gilles Sadowski < > gil...@harfang.homelinux.org> wrote: > >> On Wed, Jul 18, 2012 at 11:05:28PM +0200, Gilles Sadowski wrote: >>> Hi. >>> >>> Cf. issues 764 and 823. >>> >>> What would be the preferred class to be used as the RNG in a >>> distribution's default constructor? >> I somewhat arbitrarily chose "Well19937c". >> Please open a JIRA if this is to be changed. >> >> >> Gilles >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org >> For additional commands, e-mail: dev-h...@commons.apache.org >> >> --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org For additional commands, e-mail: dev-h...@commons.apache.org
