[computer-go] Monte-Carlo for Tactical Search

[Benjamin Teuber]

Hi Don,
I get the impression our little debate was rather based on a 
misunderstanding than a real different view...

Global search is fine, as long as it does something more than just 
playing out games until the end - of course, even that might be a 
misconception of mine.
And while others are working to improve the search in general, I'd like 
to think about the things needed in an evaluation function, and how to 
make use of MC for these things.
I also don't think the evaluation should replace MC by a 100% - they 
should rather be concurrent sources of information, whose reliability  
can be judged using probability theory.

Benjamin


Don Dailey schrieb:
> Whenever I mention the phrase "global search" it's usually followed up
> with something like,  "that doesn't work" or "that's stupid."   So I'm
> on your side of this issue.   Most people assume I mean brute force
> stupid search but I envision something like what you are saying.
>   

> I don't know if monte carlo programs will every reign supreme - they
> seem to be doing ok on small boards and I think they will be improved
> tremendously - but the way they are done now they are basically global
> searchers.   The "monte carlo" part could be replaced with an evaluation
> function.   
>   
>
>
>   
>>> Also,  it's very clear that monte carlo is very scalable.  In 
>>> real terms that means it will eventually "outrun" any 
>>> technique that is not. 
>>>       
>> This is clearly a true statement, but it seems that you are assuming that
>> the existing strong programs use techniques that are not scalable.  But
>> existing progrrams are scalable.  They are running on much stronger
>> processors today than they were 20 years ago, they use more time, and they
>> are stronger.
>>     
>
> I'm only arguing for scalable methods, but I admit that I didn't realize
> the best programs are scalable.  I should listen to the real programmer
> more - many others are very critical of using search "because it can't
> possibly produce a super grandmaster."
>   
>
>
>   
>> David
>>
>>
>>     
>
> _______________________________________________
> computer-go mailing list
> computer-go at computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/
>
>