RE: [computer-go] Pattern matching - example play

["David Fotland" <fotland@xxxxxxxxxxxxxxx>]

This is a little strong :)  see comments below...

> -----Original Message-----
> From: computer-go-bounces@xxxxxxxxxxxxxxxxx 
> [mailto:computer-go-bounces@xxxxxxxxxxxxxxxxx] On Behalf Of 
> Frank de Groot
> ----- Original Message ----- 
> From: "Mark Boon" <tesuji@xxxxxxxxxxxxxxxxx>
> Subject: RE: [computer-go] Pattern matching - example play
> 
> 
> 
> > I'm sure that the Chess and Go discipline can learn a lot from each 
> > other. But we don't speak the same language. Nullmove is called 
> > 'tenuki' or pass-play in Go
> 
> 
> Well, I guess the Go world will just have to adjust to the 
> chess world and call it Nullmove. Nullmove is a neutral term, 
> it means "not moving" (= pass) and it is "idiotic" to insist 
> on some obscure Japanese Go term when we talk about nullmoves in Go.

I think the work "pass" is a little more well known by game players than
"nullmove".  I think Mark's use of quotes around Tenuki means that he knows
this is not a term known to non-go players.

> 
> You are again putting yourself on a pedestal, underserrvedly. 
> Nullmoves are not invented by Go programmers and Go 
> programmers have not succeeded in producing strong Go 
> programs, neither have they used too many chess programming 
> tricks in their software.

We might know more than you assume :)  I don't use many chess techniques in
Many Faces
because they don't work well in strong go programs.  I use alpha-beta
search where appropriate, but my life/death search is best first, since the
trees are small, and it's easy to do good move ordering.

I understand all of these techniques, and I've used them in my other game
programs
(chess, world war chess, and Arimaa).

> 
> Therer is a massive ego-mania in Go.
> Absolutely huge ego's are involved and this is by far the 
> biggest hamperment on computer Go progress!

I feel the opposite is true.  Every few years a chess programmer discovers
go, and says "you guys are all idots. You don't use PVS/nullmove/etc.  I
know all these techniques that you don't use, so I can make a strong go
program that will beat everyone."  So far thay have all failed, because the
key to a strong go program is a strong evaluation, not better search.  The
search tricks are irrelevant without good evaluation.  There have also been
many people who said thay could use machine learning to make a strong
evaluation.
They have all failed as well.  The strong programs ALL have strong
evaluations, 
and those evaluations are ALL created by strong go players, or with
consultation
from strong go players.  Every program that became competitive quickly was
written by
a very strong go player (5 or 6 dan amateur).

So when we say you need to be a strong player, and the chess search tricks
don't help 
much, we have 20 years of experience watching people try other approaches
and fail.

I won't call anyone a moron, and it's certainly possible that there could be
some
breakthrough.  But there is a lot of history showing that strong go program
are built
by strong go players, and that they don't use PVS/nullmove/etc.

The chess programmers seem to think that go programmers are all stupid since
they can't
use chess techniques to make a strong go program.  They don't realize that
those
techniques are not appropriate for go programs.  Some of them make an
attempt at go programming
and discover this.

> 
> It already starts when someone announces to work on a Go 
> program. Then it's: "You moron, what makes you think you can 
> do what other's couldn't, first go and become a pro, then 
> piss off and study for 10 more years and have 20 years of 
> experience coding search and when you are fluent in Japanese 
> come back and play with the big boys".

There is no more need to be fluent in Japaanese than there is to
be fluent in German to code chess.  There are common game terms
that are not ordinary English word.  What would you think of someone
who said he could make a strong chess program, but didn't know the chess
meaning of "pin", or "zugzwang", or "weak bishop", or "passed pawn".  You
might advise
him to read some chess literature and learn these concepts before
he coded his evaluation function :)  Likewise, to play good go, you need
to understand things like "aji", that have no simple English equivalent.

Regards,

David Fotland

> 
> When I said I was working on a Go program I was basically a 
> persona non grata in a local Go club.
> 
> And then there is the issue with game records.
> The pros keep them secret and the people who make game 
> records available make truckloads of cash on them and 
> everybody is afraid to re-use them. There are concerted 
> efforts NOT to make game records available.
> 
> 
> 
> > We all know more effort has been put in Chess than in Go, 
> there's no 
> > question about it.  But now we see these chess-programmers 
> here and I
> can't
> > escape the feeling they think they know so much more, where on the 
> > other hand they ask the most trivial questions or make the 
> most stupid 
> > remarks that clearly show they don't know the first thing 
> about why Go 
> > programming is hard and what is involved. Well, you'll have to find 
> > out by yourself as
> I
> > don't think you'll get a lot of useful information out of 
> people you 
> > call idiots or assholes.
> 
> 
> You judge too quickly.
> It is VERY EASY for anyone (I mean a 15-year old kid that 
> doesn't play Go with a PC) to see that Go is much harder, 
> search-wise, than chess. When you accuse one of the best 
> chess programmers to be below that level, even though he has 
> obviously spent quite some time on Go (even coded up a lot of 
> code), you are simply insulting.
> 
> Any idiot sees immediately why Go is harder than chess.
> When I was a child (I was a minor) I fully understood why 
> programming Go was harder than chess. That is 24 years ago. I 
> had a PC and I was programming a compiler to be able to write 
> a chess program.
> 
> I am sure Vincent did similar things and attacking people 
> because they use "nullmove" and not "tenuki" is plain silly.
> 
> 
> > That is generally what science is about.
> 
> Bullshit.
> Science is about hard facts and not beliefs.
> If I had to work with "beliefs", I would literally have been 
> dead now, as in practice, medical people (as all people) 
> prefer personal beliefs and adapt what they read to what they 
> believe. Basically, only 1% of people is looking for the 
> truth, the rest is looking for factoids that supports their 
> beliefs. And ego often throws a monley wrench into the 
> system, as well as vested interests.
> 
> 
> > Until one of the belief systems is backed up by some sort 
> of facts or
> proof
> > there's no argument to favour one belief system over the 
> other except 
> > for personal preference.
> 
> 
> You don't need facts when you can use logic.
> And as long as you have a belief only that is not backed up 
> by facts or logic, by definition you are a fool. That is 
> namely the definition of foolishness.
> 
> 
> > Finally, to rub it in, I believe the brain is nothing but a 
> > sophisticated computer.
> 
> This is a very general statement.
> The brain is a massive neural network with digital neurons 
> (pulsetrains) and (chemical - infinite) associative memory. 
> It is as far from a conventional (von Neumann) computer as an 
> ant is from a whale. You are saying: "I believe an ant is an 
> animal just like a whale is an animal".
> 
> 
> > I also believe humans play Go by doing almost nothing but 
> local search 
> > combined with pattern-matching and whole-board evaluation. I
> believe
> > humans play Go well. THEREFORE, I believe you can make a Go program 
> > play well almost solely based on local search. Anyone who believes 
> > full-board search is the ONLY solution is obviously mistaken.
> 
> Vincent has already clarified his position and you are lagging behind.
> 
> > Moreover, according to my
> > beliefs, the possibility of a good program based on local search has
> already
> > been proven, whereas the possibility of a good program based on
> brute-force
> > has not.
> 
> I was not aware that there are any good Go programs?
> You think 8k is good?
> 
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx 
> http://www.computer-go.org/mailman/listinfo/computer-go/
> 


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/