I don't think this specific test had been done. But I'm assuming the result will be the same as previous tests: deviating from <the pursuit of the the highest
winning percentage> leads to a degradation in strength.
Brett Koonce wrote:
Greetings from a lurker,
Forgive me if I am talking out of my hat. It has been a long time since
I have done any real coding.
It seems most of the gains in MC/UCT come fairly quickly (or rather you
can get within 50% of a good move guess with a few iterations). It
would be interesting to perhaps do a progressive stepping down/widening,
i.e. 1k playouts with komi + 3 as the cutoff, then feed this tree into
2k playouts with komi + 2, then 4k playouts with komi + 1, and then
finally do the usual full blown regular analysis, say 50k playouts
(numbers can be tweaked of course). You would lose the initial
simulations from your final one, so you would be sacrificing say 10% of
the possible simulations, but on the other hand it would seem to bias
the tree toward making moves that have a greater chance of winning by a
greedy amount without explicitly telling the computer it has to win by a
certain number, which would seem dangerous if the simulations are near
the threshold.
I apologize if this is an obvious idea, was just wondering if there was
a flaw with it/someone had done experiments in this direction already.
-Brett
_______________________________________________
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/
