Joined: 30 Dec 05
Was reviewing the results gathered by dcdc, discussed here. His first chart happens to report the random number seed used for each work unit.
I understand that a seed is used in a random number generator (RNG), which is basically a formula which uses that starting seed and produces some unique sequence of random numbers. The stream of random numbers produced are used as a starting point for each model run by my machine. Some tasks and machines will use just 1 or a few random numbers from the stream, others will use a 100 or so.
I notice from the list that the seeds assigned to tasks are numbered sequentially. It would seem to me that a better overall distribution of generated random numbers would be produced if the seeds used were random as well. Especially since we are collectively using only a small number of the possible seed values in the work that we do for Rosetta.
Is there anyone that knows enough math to confirm my suspicion above?
Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
Joined: 17 Sep 05
You may be correct. However the number generated from seed and seed+1 will normally not exhibit a noticable pattern. So the number of seeds should be more important than the relationship between the seeds for producing good results.
BOINCing since 2002/12/8
Joined: 18 Sep 05
Using Visual Studios probably crude random number generator to generate 10 integers between 0 and 10 inclusive with a seed of 12345 gives:
3 6 8 7 2 8 9 2 9 8
Using a seed 12346 gives:
3 10 4 5 5 9 10 1 4 6
A decent twister or better would be even better.
Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.
©2023 University of Washington