On Monday 19 May 2008 07:51:22 pm ddneilson wrote: > You're over-thinking the problem a tad. Seeding the random number > generator via the loop counter from your cluster-submit script is just > as random as anything else that you might come up with; as long as each > program has a different seed, then they'll each use a different sequence > of random numbers. seed(rand()) may feel more random, but you still have > to come up with an initial seed or else you're just going to end up with > the same random sequence in each program. > > Alternatively, if you don't mind throwing the ability to duplicate your > results exactly, you can always seed with the results from `date +%N` -- > which will output the number of nanoseconds since the last second. i.e. > you can make the line that runs your program, in your script, into > "./myprogram `date +%N`" (those quotes around the date call are the > forward quote on the same key as tilde [~]) and seed from the first arg > of your program. > > -Daniel
Yes, indeed, I was thinking of putting something other than just an ordinary loop parameter as a seed (indeed, it "feels" more random to put rand() there), but I agree it doesn't change anything. The idea with microseconds is really nice, but I think being able to reproduce data is a very nice thing I haven't thought about :-) Thanks. Sergey _______________________________________________ Help-gsl mailing list [email protected] http://lists.gnu.org/mailman/listinfo/help-gsl
