On Mon, Mar 31, 2008 at 03:12:39PM -0700, Christoph Birk wrote:
>
> On Mar 31, 2008, at 1:05 PM, Don Dailey wrote:
>>
>>
>> Christoph Birk wrote:
>>>
>>> On Mar 31, 2008, at 10:48 AM, Mark Boon wrote:
>>>> I don't know about this. I'm pretty sure MoGo checks if the stone can
>>>> make at least two liberties (ladder problem) in which case it can
>>>> still be horrible but very seldomly worse than random.
>>>
>>> I would expect playing a "not-working" ladder to be worse than random
>>> most
>>> of the time.
>> Of course this is true, but presumably a move that answers a direct
>> atari threat would classify as being better than random.
>
> Not if it's a working ladder.
I think this is not obvious, since there is still about 50% chance on
who gets the second move there. The original MoGo pattern describes only
capture-possibility check, not atari-possibility check, so even if you
play out the ladder, the opponent will most likely tenuki. So I think
not playing out working ladders as white is actually better because it
saves _black_ from getting a bad result!
I implemented a naive ladder check (that does not catch a lot of cases,
especially when you have to decide whether to "slide" the stones in
ladder along border or some of your friendly stones) and tested its
efficiency in UCT-driven playouts with only a 50% probability
capture-possibility check heuristic.
Against GNUGo Level8, I get 37% +-5.7% wins with ladder check, 28%
+-4.6% without ladder check. Unfortunately, the numbers are not very
precise and the difference is probably much smaller in reality - 37%
seems overinflated given later measurements; I will redo the test with
newer codebase and more games.
(I plan to post detailed measurements of effects of various heuristics
and parameters anyway; so far it seems my code is still too buggy
though: AMAF, prior knowledge and FPU all make my program weaker :-(.)
--
Petr "Pasky" Baudis
Whatever you can do, or dream you can, begin it.
Boldness has genius, power, and magic in it. -- J. W. von Goethe
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
