Spinroot

srirajpaul

Member

Registered: 2013-09-07

Posts: 13

Randomized search coverage

Hi,

I have a model[1] which is quite big because of which I am doing randomized search instead of complete one.
The hash factor comes to be around 2. We are trying to get more coverage be running it with random seeds.
Is there any pointer to how many random seeds need to used to get some good coverage?
Also is there some guidance in the selection of the seeds so that the state space explored don't overlap so much?

Thank you
Sriraj

[1]:http://srp7.web.rice.edu/new_pe.pml

spinroot

forum

Registered: 2010-11-18

Posts: 702

Website

Re: Randomized search coverage

There's no guidance other than that each random seed gives you a randomly different sampling of the statespace.
For a hash factor of 2, the true statespace could be several orders of magnitude larger than what you've sampled so far.
Short of moving to a larger machine, with more RAM, all you'd be able to conclude is that each additional run with random seed improves your coverage.