Re: Seeding the random number generator

["Randall C. O'Reilly" <oreilly@grey.colorado.edu>]

james@let.rug.nl writes:
> I recently did some training runs in parallel on a linux cluster only
> to find that I got identical results in each run. Could this be down
> to the processes being started at the same time?

Are you doing ReInit at the start of training?  You should do a
NewInit in any case to ensure a new random seed.  This is indeed
triggered from the clock, and if in fact you are doing NewInit in this
way and your clocks are so exactly syncronized then perhaps you could
share the alternate technique that doesn't suffer this problem?  The
relevant code is in src/ta_misc/random.h,.cc

				- Randy