Vlad Dumitrescu wrote:
> Hi Don,
>
> On Dec 10, 2007 9:08 PM, Don Dailey <[EMAIL PROTECTED]> wrote:
>   
>> /snipped a lot of interesting stuff/
>> However MC play-outs is not horizon limited like this.   It's stupid to
>> make it greedy because it may notice that winning the big group leads to
>> a loss every time and that some other course of action is more productive.
>>     
>
> I like the way you make your arguments. I have a question about the
> above, hopefully you or somebody else will know the answer.
>
> Am I confused in my understanding that a weakness of MC evaluation is
> that due to its random play it will miss sequences where there is only
> one winning move at each play? 
Random play-outs of course is flawed - even an infinite number of them
will not return the correct score.   A common example is that an
unmodifed random play-out is prone to making atari moves even when the
stone making the threat is in atari.    If you attack a big group with
one stone,   on average you will come out ahead even if you lose that
stone half the time.     So it's kind of a lottery,  there is a "good"
chance you will kill this group before he kills your attacker,  or that
you will accidentally defend your attacker before he captures your atari
stone!    It doesn't work out statistically to the right value.

> This is the way I am interpreting the
> "nakade problem" discussed in another thread: to keep a dead-by-nakade
> group dead, one must not miss one single move in the sequence. Of
> course, in games the nakade example is just one of the simpler
> variants, most semeai (capturing races) will fall into that category.
>   
Well you also have the UCT tree part  which methodically repairs the
problems.    My understanding of the Nakade class of problems is that
the play-out is biased in such a way that this is not easily seen.   A
random search may be better in this specific instance. 

> This uncertainity is what gives the less-than-1 confidence you
> discussed, but my feeling is that it varies too much with the sequence
> length -- the answer would be to add some intelligence, like MoGo and
> the other top programs do.
>   
Yes.    You may not have 100% confidence, but it's a huge step in the
right direction.   After all,  how can you do better than actually
matching the goal of your program with the actual goal of the game?  
Beginners essentially make it their goal to win stones, advanced players
set the goal of doing whatever it takes to win.

I'm a pretty lousy player still,  but I started out trying to capture
everything I could.   It eventually occurred to me that is was far
better to defend 41 points that was defendable than to go off on hunting
expeditions and it was far easier to think about the game.

- Don


> best regards,
> Vlad
> _______________________________________________
> computer-go mailing list
> [email protected]
> http://www.computer-go.org/mailman/listinfo/computer-go/
>
>   
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/

Reply via email to