RE: [computer-go] Pattern Matcher

[Vincent Diepeveen <diep@xxxxxxxxx>]

At 16:55 7-11-2004 -0200, Mark Boon wrote:
>> -----Original Message-----
>> From: computer-go-bounces@xxxxxxxxxxxxxxxxx
>> [mailto:computer-go-bounces@xxxxxxxxxxxxxxxxx]On Behalf Of Vincent
>> Diepeveen
>> Sent: Sunday, November 07, 2004 12:04
>> To: computer-go; computer-go
>> Subject: Re: [computer-go] Pattern Matcher
>>
>>
>> Please don't act overarrogant. I have met past 15 years enough arrogant
>> computer game researchers to recognize one when i see one. In
>> fact i'm sure
>> without arrogance and stubbornness one can't write a program of his own.

Go computer world so far has been saved from those thousands of researchers
with clever ideas on how to do automatic learning.

99% of them is not busy with recognizing patterns IMHO. You can deduce
their 'learning' to parameter optimizations of a known program.

Now very good tuners such as Chrilly Donninger (IMHO worlds best program
tuner), can really get a lot out of a program by tuning it better. 

We have had a very nasty case in computerchess called 'TD learning'.

My friend in the other chat window is one of worlds #1 experts on ANN's.

He had a lot of problems in which the experiments with TD learning were
performed. Yet all those researchers share 1 thing. They are good in
creating noise and looking better than they are, without achieving anything.

They never have a program that wins tournaments and they never show
anything which has a statistical meaning.

In computerchess nowadays there is many scientists who require more
statistical proof. That has caused majority of them to leave.

Note that i find it very good such research gets done.

At this moment i'm helping a master thesis student who is inexperienced in
optimizing software in order to let bigtime searching software calculate
statistics and hard data about proteins. Even for this research he's
already going to do 23000 expensive long lasting calculations in order to
get more data.

If you see in such research areas like health and environment, what
statistical significance they must show before something is accepted as
being the truth, then in game tree search we really are very soft towards
many researchers which do not show any logfile nor do any objective
comparision, let alone show statistical significant results.

Frank de Groot can claim anything of course if he never is going to join
tournaments with it.

At least in the computerchess some have been more clever than him.

There they write sometimes a good program and then win a tournament and
then claim anything in the skies. 

>This is very true, and I'm not free of such arrogance and stubborness
>either. I agree that to a certain extent it seems to come with the drive and
>determination to be involved in a project like this for years. Maybe my
>remark that triggered this outburst of Frank wasn't very nice, I just
>couldn't help myself for a minute there seeing such a silly statement. But
>I'm also not a hypocrite, and I don't continuously hide my real opinion
>behind niceties.
>
>I couldn't help noticing that chess-programmers seem to be much more
>aggressive than Go programmers, and the amount of nastiness and mud-slinging
>is far greater. Maybe it's because the stakes are much higher? Then again,
>the same holds true for real play. The amount of fuss that I occasionally
>hear about that happens in (professional) chess tournaments would be
>unimaginable in the Go world. And at least at pro level it's definitely not
>true that the stakes are higher in Chess. Interesting stuff for an
>anthropologist I'd say.
>
>> But please don't do as if i know nothing from patterns. I work fulltime on
>> patterns. Not so much in go, but in chess.
>>
>> I happen to know exponential more than you there about the advantages and
>> disadvantages of patterns.
>>
>> Note that my religion is Christian. My religion isn't condemning automatic
>> generated patterns used either for move selection or evaluation.
>>
>> In your completely wrong and naive calculation below you forget 1
>> important
>> thing which all the automatic approaches on games seem to forget.
>
>I have beliefs, but I'm willing to change those beliefs when they're
>scientifically proven wrong. Which I suppose by definition makes them
>non-religeous. My belief, as seems to be yours, is that this
>pattern-generating approach is not going to work to make a strong Go
>program. But it doesn't help saying to each other things that implicate
>something like 'you don't know the first thing of ...' (fill in the blank).
>
>Let's have the facts do the talking. So far I have seen no relevant facts to
>support either the claim that it will or won't work. So I stick to my belief
>that it won't work, as I think I have very good reasons for them. Despite
>Frank continuously claiming he already explained everything, I get the
>feeling he's only giving out information bit by bit. My hunch is that this
>is because he thinks this information is too precious to share with anyone.
>But on the other hand occasionally the urge to boost his ego by bragging
>about it here is also very strong and gets the upper hand. This is only
>natural, and I don't want to blame him for that. I consider this in the same
>category as the arrogance and stubborness mentioned earlier. What I do mind
>a bit are his unfounded claims and speculations that are not backed up by
>any verifiable data. Not to mention the fact that he gets nasty when someone
>disagrees with him.
>
>If there's any worth in Frank's claims then we'll see proof of it soon
>enough. In this sense this is an easy topic when it comes to proving who's
>right. I get my program, you get yours. If your program wins then your ideas
>were better. And it doesn't matter if the other was nice or nasty. Other
>than that, it's all pure speculation.
>
>
>
>_______________________________________________
>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/