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

[Vincent Diepeveen <diep@xxxxxxxxx>]

>From Frank de Groot>I wish someone would peer-review my sourcecode (as done
in XP). Every time I get better results I am convinced I finally "did it"
and can
>move on to the next stage, now I just discovered another huge bug that
>necessitates re-crunching for a week.

>It turns out that my statistics of the harvested patterns are based on
>approx. 10 times less data than I thought, so only on 50,000 games. The good
>news is of course that the performance will go up again but I wonder what is
>wrong with my brain..

Wow what a timing this posting...

I read this just at the moment i was drinking something,
it's on the floor now :)

>
>> have obtained.  It's nice to see claims backed up by such solid research.
>
>Of course.
>But the problem is with commercial Go programmers that firstly it will cost
>them a lot of time to publish, secondly they might have great difficulty
>publishing anything (I for example do not understand mathematics), thirdly
>it is totally against the interest of commercial Go programmers to publish
>anything.

The only problem with publishing official articles in ICGA for the
programmers is that you don't get paid for it.

On the other hand i write on average 500 pages of reports a year at least
about DIEP for different magazines and websites.

A problem i have with most articles written in certain journals is that
there is zero statistical significant data inside it.

Even to measure a single parallel speedup i already run 213 positions
automatic at the program in order to measure whether it works. None of
those article writers ever does and they have ALL DAY to do it.

>So I think when a commercial Go programmer like David Fotland or Mark Boon
>actually share ideas, principles and even source code (I am not sure Mark
>Boon is still making money with his most expensive Go program (around 150
>USD?), that this as only laudable and it's not really fair to "demand" a
>publication.

One tip: never ask people who sell software how many copies they sold.

You may know what i sell currently from diep: 0 copies.

I don't sell anything of it right now. So i have zero pennies to pay my
breakfasts.

>I have never understood the reasons for extensive proof anyway. Just read

No one demands a proof. Just statistical data to backup something works.
What i do with all articles describing something new is first browse to the
last page of the article and see how many tests have been performed. If no
table is there, i skip already the article.

Doing that i by accident skipped article of a good friend of mine previous
magazine :)

I was done with the magazine within 30 seconds this time, noticing an
olympiad is there start of september 2005 in Taiwan.

>the concise explanation of the idea, if you think it will work, use it.. My
>pattern system is based on a publication. But since I do not understand
>mathematics I only looked at the pictures <g> and I didn't care about the
>results either! I thought "This is a cool idea and I can make this work".
>That's all. Inspiration.

Results matter. In computergame world regrettably you can get categorized
in 2 categories. Either 'winner' or 'loser'. And winners take it all...

Note i'm not like Amir Ban. He always says: "winner is always right".

>If a programmer tells me: "I did such-and-such and I have these amazing
>results" and I think he's just a clown and his explanation of his idea looks
>like the rantings of a madman to me I am not going to ask him to publish
>anything because that's basically saying: "You are a liar, I don't believe
>you". What difference would a publication make?
>
>Neither would I ask a publication if his story/results were credible because
>that would be saying: "I believe you but I do not understand your
>explanations and I do not understand the explanations of your explanations
>and no amount of comp.go list discussion makes me understand it more, but I
>would like to rip off your work anyway so please spend a month of your time
>to hand over all your intellectual property to me".

>Researchers publish stuff to establish something, commercial programmers
>keep things secret. When a commercial programmer dicloses anything, it

I learned a lot in my life from researchers who after a few years quit the
computer game world and are not having problems to tell anything about what
they did.

Many of them have very good ideas and realize very well what is the problem
with your program and what is important for the future, or just have real
good ideas on how to do certain things.

>means: "I want to cooperate in the community of peers = fellow COMMERCIAL
>programmers". When fellow commercial programmers remain silent about their
>latest developments or "secret" stuff, of course it was a mistake to explain
>about the internals of one's own "secrets", but that is a chance one takes
>only once.

just remember one thing. publishing in nerd magazines and posting at
mailing lists eats time, but it never gives you money.
_______________________________________________
>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/