Author: Vincent Diepeveen
Date: 06:20:59 12/12/02
Go up one level in this thread
On December 12, 2002 at 09:08:25, Maurizio De Leo wrote:
>You seem to not want to understand what we are talking about. The question was
>
>>>>>WESTERN CHESS PROGRAMS ARE GM-STRENGTH.ARE GO PROGRAMS EQUALLY CAPABLE?
>
>And you answer was :
>
>>>>If similar effort would have been put in go like it has been in chess,
>>>>my answer would be YES.
>
>You also write :
>
>>The difference is that a single person will be capable of writing a
>>go program within a year or 2 that can easily challenge the world top.
>>In my case if i would be busy a full year fulltime, i would surely,
>>even without help of a strong go player, be capable of challenging the
>>go-top.
>
>I don't doubt your programming skill, but the question wasn't if you can
>challenge the top computer-go program of today (which are pathetically weak). It
>was if you can make a go program that is Gm (= professional) strenght. And the
>answer is NO. Not with today hardware, not with today technique.
>I showed you. You claim that the branching factor of go can be reduced to 10
>ply. Also if that is true, which I doubt, there is a HUGE difference between
>10^x and 4^x !
You showed nothing.
You just show you know nothing how far the chess programming world is.
You didn't investigate any go program i bet.
Yes i wrote within a few hours a go program a few year ago.
With that thing i could already search 6 ply at very old
hardware. No problem.
At the goprogramming mailing list they still are busy confusing
local and global searches with each other instead of a single
global search like we do in chess already from 1980 and on.
It seems you have very little knowledge here. The 1980 programs
which did such a stupid form of search were very idiotic.
The main problems for GO is:
- you can't sell your stuff easily, because the
vaste majority of buyers live in Japan (china is not
relevant, they earn too little to afford a go program; so
they copy it i guess).
The Japanese market is very closed
- you need to invest time to do it. Being myself a titled
chess player it is not so hard to imagine that i prefer
putting my time in computerchess rather than computergo at
the moment.
- the current programmers are searching very fast locally
but globally they do not know how to program at all. For example
i was amazed that in a time where my go program searched at
3000 nodes a second that they couldn't even get 50 nodes a second
at the same hardware. Yes my thing did calculate influences of
groups and it did take into account which groups were hung and which
weren't. That was just an afternoon of work of course. Evaluation
was based upon the Goliath paper describing how to evaluate which,
memory says it was written in dutch, was of course very well readable.
You see a factor 60 is pretty much difference in speed.
My current machine is a dual 1.6ghz MP. Ten times faster at least than
the machine i had back then.
A big problem of Go programs when i test them is that when you give
them more time that they do not play better, some even play worse
moves.
That's of course ridicioulous. Don't you agree?
It is a simple matter of amateuristic programming, but all computerchess
programmers exactly know what causes it. Somehow in the computergo
mailing list they do not.
Best regards,
Vincent
>Down here there is some mumbling about the current computer-go top, which isn't
>on the topic
>>However it is impossible to write a chessprogram within 1 year that can
>>challenge the world top computerchess. *impossible*. Even the best
>>go programmer will be capable of telling you why.
>>They are just busy figuring out knowledge from chessprogrammers and
>>which is known in the computerchess world for years, in order to improve
>>their own go program.
>>They should figure out things like *how* do some of the oldy chessprograms
>>do their selective search?
>
>
>>>Ok. So let's analyze. For 6 ply 10^6 moves in Go and 4^6 = 4096 in Diep (or do
>>>you have a branching factor over 4 ?). It seems that you have to squeeze some
>>>three order of magnitude from optimization.
>>
>>Every beginner can get a better b.f. than the top go programs, but
>>you won't be capable of writing a correctly searching chess program that
>>can challenge the strong chess programs b.f.
>
>So what ? Read again the starting question of the topic. I wrote it in bold, at
>the top of the message.Let's assume "every beginner can get a better b.f than
>the top go programs",as you say.Anyway there is no chance to write a GM strenght
>go playing program, and that is the point. If top go program have branching
>factor of 12 and you claim to achieve 10 or 8, this still will leave you WAY
>behind chess programs.
>
>After this there are some question which you didn't answer at all. I numbered
>them, so you can more easily answer.
>
>
>>>>Add to that, that there is no GM title in GO.
>>>
>>> (1) So ? What this mean ? (1)
>>>There is no "professional" title in Chess. There is no "Davis Cup" in soccer.
>>>And so on with nonsense statements.
>
>>>>Suppose you let a small FM play against a strong chessprogram at a level
>>>>of a full day for the entire game.
>>>>Of course the FM will win.
>>>
>>>What ?
>>>Usually go players have 6/8 hours for the game which isn't far to 2h/40+2h/40
>>>+30min of top chess. Anyway you are a strong Fide Master, almost
>>>international master.
>
>(2) Would you mind beating Deep Fritz 7 by 10-0 (i would accept also 9-1) ?
> Maybe time can be 10 hour for player for the game ? Or 12 hour, as you
> want. (2)
>
>
>Maurizio
This page took 0.01 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.