On Jan 26, 2009, at 6:26 PM, matt harman <harman.m...@hotmail.co.uk>
wrote:
> That the missunderstanding right there.
> 1 child will be chosen and 1 simlation will be run.
Thanks for the quick answer, so 1 simulation is run because too many
will give lots of noise to the result?
Just the opposite. The noise in the win rate is 1/sqrt(n). The reason
for doing few simulations is that they're relatively expensive. UCT
and MCTS do extra tree walking because that's less expensive than over
simulating bad moves.
if only 1 is run then the 4 children can either win or lose
the single simulation 0 or 1.
Initially, yes, but future visits will increase that. Near the root
could have hundreds of thousands.
This would be non-deterministic so how would you
decide which child to exploit?
Techniques vary a bit, but a move quality is calculated based on the
simulation results for each move. The best move is simulated.
UCT balances exploitation with exploration by including uncertainty in
metric.
Many top programs have discarded the uncertainty term because
heuristics and RAVE/AMAF are surprisingly accurate.
Given specific simulation results, the searches are deterministic.
Simulations are random, so the searches end up very non-deterministic.
Over time, they should converge on the same conclusion.
Thanks
Matt
> _______________________________________________
> computer-go mailing list
> computer-go@computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/
Beyond Hotmail - see what else you can do with Windows Live Find out
more!
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/