Re: [DISCUSSION] Game Tree vs. Pattern Matching

[Mousheng Xu <xu@xxxxxxxxxx>]

At 10:04 AM 10/29/98 +0200, Antti Huima wrote:
>On Wed, 28 Oct 1998, Mousheng Xu wrote:
>
>> Dear Smart Guys,
>> 
>> 	I am having a hard time in deciding whether I need to build a
>> game tree or a pattern matching database. Here are the comparisions of
>> a few aspects between game tree (assume min-max) & pattern matching.
>>
>> 		GAME TREE		PATTERN MATCHING
>> Space:	much smaller		much bigger
>> Efficiency:	log(n)			log(n)
>> History:	sensitive		insensitive
>> Time:		dynamic		static
>> 
>
>Ignoring that, I think you cannot make a choice between game tree search
>and pattern matching. A program playing using only local and relatively
>static pattern matching will not become strong. By 'local' and 'relatively
>static' I mean that in principle a 'pattern matcher' can incorporate
>arbitrary tactical searches as a part of the matching process; it's just
>about how you defined your 'patterns'. But with 'conventional' patterns
>alone, success will not be achieved, because non-local properties as well
>as all move continuations are ignored.

	I surely agree with you. An good evaluation function is almost all we need
to make a good go program.
I have been working on a "good evaluation function" for years. I feel it's
time for me to crank out a whatever ugly/beautiful, weak/strong program,
and then continue working on evaluation. A game tree or pattern database
will save some calculation time.
	Really thanks for your comments.

-- Mousheng Xu