t (only 1 move = win).
Do you mean you found some problems are wrong?
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
see why D2 works. After E2, can black live?
>139: C2 and B1 also live.
Yes, but will not win with komi 7.5.
>143: I don't see how A3 could win the semeai. A2 and C4 look more
>effective.
Typo, it was A2. C4 cannot work.
--
Yamato
___
Go could run on the MC only mode, it might be more
interesting, I guess.
Gunnar Farnebäck wrote:
>One more correction. The fixed version added B1 as a correct move in
>113, but that point is occupied.
My mistake. There will be no effect on the results, anyway.
--
reg_genmove black
#? [B1|D1]
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
John Fan wrote:
>On Problem 125, sounds like B2 is also a right move.
B:B2, W:B1, B:C1, W:A2, B:A1, W:C2, B:A3, W:D1.
Black seems to cannot win this ko.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.
Magnus Persson wrote:
>The program reivax on 9x9 CGOS seems to be strong but suffer from a
>bug leading it to pass too early, and thus it often loses games
>against weaker programs that do not resign.
Who is the author of Reivax? It's like a CGOS deflat
/ 5817 (64% win)
D7: 2195/ 3378 (64% win)
E3: 918/ 1742 (52% win)
C6: 863/ 1707 (50% win)
C5: 779/ 1418 (54% win)
It seems the best move is D6 or D7.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
> I ran 86 19x19 games with both on the same hardware (single core of
>A64 X2 6000+, 2G RAM) with 20 minutes S.D. each, the rate is MoGo win
>83.3% (+-4.1).
How did you set the time to 20 minutes S.D.? MoGo doesn't update the
clock if you don't send time_left, and Fue
time settings completely this way.
>I will rerun it with --totalTime 1200.
I suppose that --totalTime also doesn't work if you use twogtp.
--time will work.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
oblem
with UCT tree.
Look at the attached file. This position is win for black in Japanese
rules, but the only correct move is pass. If black plays anywhere other
than pass, he loses. This time white's correct move is pass, otherwise
he loses. Such a condition breaks winning rate values in
your program choose pass correctly?
>You say "white's correct move is pass, otherwise he loses". That's not
>entirely correct. White loses no matter what he does, whether it's a
>pass or anything else.
I meant that for when Black plays wrong move.
--
Yamato
Rémi Coulom wrote:
>Martin Mueller wrote:
>> Zen has been getting very impressive results on CGOS. Yamato-san,
>> could you tell us a little bit about yourself and your program?
>>
>> Thanks
>>
>> Martin
>
>Yes, very strong results. Congratula
You are right, I was not aware of the KGS rule.
I think I have fixed the problem now.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
ill be a commercial product but currently I have no specific
plans. I don't have plans about a massively parallel version either.
Unfortunately I cannot participate in the Olympiad. For KGS, I can.
--
Yamato
___
computer-go mailing list
computer-go@comput
more important than mine. If you don't
use minorization-maximization yet, I recommend studying it.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Ingo Althöfer wrote:
>Other question: Might it be possible to find a
>volunteer for operating Zen19 in Pamplona?
I forgot to say it here - Mr. Kato came forward as an operator of Zen.
It will play via KGS. Have a good game, all.
--
Yamato
___
co
Carlo search with no
>tree on small boards.
I like you idea, but why do you use only 5x5 and 6x6 Go? I don't think
the 200+ Elo improvement is so impressive because the previous approaches
were not optimized for such a small boards. I'm looking forward to your
results on
t's always much
>better to compare multiple implementations of the same algorithm.
Could you give us the source code which you used? Your algorithm is
too complicated, so it would be very helpful if possible.
--
Yamato
___
computer-go mai
pattern actually
>occurred after move a in position s) and the expected feature (the
>average value of the pattern, weighted by the probability of selecting
>each action).
I still don't understand this. Is it the formula 6?
Could you please give me an example like the above?
--
Yama
or less than we could have expected.
I understood what psi was. I am not sure how it works, but anyway I can
see your algorithm now. Thanks.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
go. (The game against MoGo was lucky, not good example.)
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Ingo Althöfer wrote:
>Congratulations to Zen author also from me!
Thanks for your all wishes.
>As far as I know, Yamato is only a nickname and not the true name
>of Zen's author. If this is the case, when will the identity of
>Zen's author be revealed? Did (or do) the orga
eki stage. Zen already has the opening
book and some joseki patterns. However, the problem is, a good position
for a human is not always good for a program, because of their feeling.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
disposal is a dream of many
>years!
Sorry to keep you waiting. Unfortunately the project is still in the
planning stage. I cannot say when or how much it will be.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
's well know that we easily find
>patterns where they don't exist too.
I have done some experiments. Zen won 30/100 games against a mechanical
mirror-go program without any anti-mirror code. The 2 continuous losses
were not unlikely, but I think there is no proof that Zen was far weaker
agai
meien.
Have fun.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Ingo Althöfer wrote:
>Yamato San wrote:
>> The match of O Meien 9p vs Zen will be held on next Monday via KGS.
>>
>> Schedule (JST)
>> Date: August 10
>> Time: 14:10
>>
>> Handicap
>> 9x9: Zen is black, with 0.5 or 3.5 komi (?)
>> 19x19
setting was 15 seconds per move.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Isaac Deutsch wrote:
>Did you think the bot would not benefit much from additional time usage?
I'm not sure. The fast time setting was the wish of the sponsor. The games
were played on the real board and they didn't want to take a much tim
My experiments might not be enough, but I feel this kind of idea is not
a solution for the handicap problem. Did anyone succeed with it?
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
rograms become weak because
of the instability of the simulations.
I think that we need a new algorithm to combine the UCT information into the
playout policy. I have already done a lot of experiments in this area, but not
yet succeed. However I believe this idea will be the ne
success.
String criticality seems a nice idea, but how should it be implemented?
Just giving high priority to the liberties does not work, because that
cannot be distinguished from the simple dame-filling.
Can you suggest a concrete formula?
--
Yamato
pty_when_string_captured / string_captured_count
>
>value 0: perfect defense
>value 1: perfect offense
Well, it should work theoretically but seems very expensive for the
processor. When we have this information, do we have to update every
node value for ev
th Go. (Tencho = Zenith)
It will be available on the Internet in the near future.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
cial version is KGS 2d :-)
Its 2-dan is equivalent to ZenLv6, weak KGS 1d.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
E, but the
results were worse than normal RAVE.
While RAVE is a very efficient algorithm, it strongly limits scalability
of the program. It typically makes a fatal mistake in the position that
the order of moves are important. We definitely need to improve RAV
played A, B is preferable for me.
otherwise it is not."
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
andidate
>move
I don't understand what it is useful for. I think that the near points
from the previous move are already preferred by proximity heuristics.
Could you give us an example?
--
Yamato
___
computer-go mailing list
computer-go@compute
widening. So what do you use?
It sounds like a ManyFaces' secret. Could you tell us how much Elo
rating you gained by this invention?
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
hers seem to have much more complex schemes. But I
>don't think I implement Rave like others do either.
Your implementation must be very different from mine. Actually I don't
use Progressive widening (or unpruning) at all. It's a mystery
ith few exceptions.
>How many playouts per second do you do per CPU on 19x19?
It is about 1000 playouts per second with 1 core.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Hello All,
I have found this slide on Google.
http://www.lri.fr/~sebag/Slides/Venice/Kocsis.pdf
However I cannot find any explanation for it.
Does anyone know what Discounted UCB is?
Is it useful for MC Go?
--
Yamato
--
Start Yahoo! Auction now! Check out the
Thank you Sylvain, your explanation is enough for me.
--
Yamato
>Hi Yamato,
>
>> However I cannot find any explanation for it.
>> Does anyone know what Discounted UCB is?
>
>"Discounted" means you forget somehow the past. More precisely, if "w"
>is
site, please inform me!
If you want, I recommend Download.com.
But I would rather you put it on CGOS or KGS.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
e a question. In section 4.1, you wrote:
> Contiguity to the previous move is a very strong feature (gamma = 23)
Where has the number 23 come from?
--
Yamato
--
Easy + Joy + Powerful = Yahoo! Bookmarks x Toolbar
http://pr.mail.
have been run." How did you determine these numbers?
Thanks
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
an for "the other way round?"
>All experiments (except the default policy) were played against GnuGo
>level 10, not level 8.
That's surprising news to me...
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http:
on't see why it is so surprising, I guess it does not change
>really the level of gnugo.
Because I have tested my program against GNU Go level 8 to compare the
winning rate with MoGo. Now I see surely there is almost no change.
--
Yamato
y using prior
>knowledge.
Sorry, what is AMAF?
And I have another question; Don't you use Q_RLGO anymore?
If so, would you explain the detail of the Q_MoGo heuristic?
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
e "a" in position "s", then 1
>- else 0.5
Thanks for that. These values were more extreme than my expectation.
I thought you use values like 0.4 or 0.6.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
>pseudo-mathematic notation.
>Why is this inappropriate notation still used?
I agree that the pseudo-code is easy to understand.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
t_rave = beta * rave_value + (1 - beta) * uct_value;
>>
>Thanks for the translation. The only point I am still missing: What is
>RAVE(AMAF)?
RAVE is another name of AMAF(all moves as first) heuristic.
The details of AMAF are explained in this paper.
http
lvain said that good moves in the playouts do not always improve
the performance of UCT. What do you think about this claim?
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
t be a very good method.
GNU Go often cannot respond to an obvious bad move correctly, so
pruning such moves decrease the winning rate.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
and in games are different things.
When MonteGNU is published, it will be an alternative of GNU Go.
Of course MC vs MC may have some problems, but at least it is
stronger than GNU Go on 9x9.
And, if the way to combine UCT and the local tactical search is
discovered, the regression test like GNU Go will
ws version as soon as possible.
Have you tried Visual C++?
http://msdn2.microsoft.com/en-us/express/aa975050.aspx
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
>> Have you tried Visual C++?
>> http://msdn2.microsoft.com/en-us/express/aa975050.aspx
>
>The thing is that VC++ does not have the pthread library.
This library might be help.
http://sources.redhat.com/pthreads-win32/
# I have not used it,
Sylvain,
Can I set the number of the simulations per move instead of the
thinking time? (like "--simulations 3000")
If possible, it is very useful for the benchmark.
--
Yamato
___
computer-go mailing list
computer-go@computer-g
>Yes you can:
>--nbTotalSimulations 3000
>
>Once you set this option it ignores all other time settings.
Thanks! It works perfectly.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mail
on
method, however we need very long time to find it.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Do you have any data to compare the performance of biased playouts
with MoGo-style one? (the winning rate against GNU Go, etc)
Also, how large and how many are your patterns?
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
his is no longer using MoGo's hand-coded patterns,
>but still uses the explicit preference for captures/saves and local moves):
>84%
Thanks for interesting results.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
s from the
Bradeley-Terry Model. greenpeep uses similar patterns extracted from
the offline self-play.
Then I have tested a lot of change of probability distributions, but
it was very hard to improve the strength.
Any comments?
--
Yamato
___
computer-go ma
testing policies is very time
consuming. I think at least 1000 games that use 3000 or more playouts
per move are needed to judge whether a change is good or bad.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
The whole process is a bit of a
>black art, not just the play-outs.
I fully agree.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
but also
>
>Zen-0.9 2386
>Zen-1.0 2385
I think it is CGOS deflation.
The evidence is:
ControlBoy 1546 8/8 100.00 <- Zen-0.9
ControlBoy 1460 5/5 100.00 <- Zen-1.0
--
Yamato
___
computer-go mailing list
computer-go@computer-go.
Don Dailey wrote:
>But FatMan is still 1800!I wonder if FatMan improved causing the
>deflation? :-)
Don, why don't you use MoGo as the second anchor?
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.comp
is formula.
beta = 1 - log(m) / 20
I think it is a little better than the one on the ICML paper.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
, can anyone write the pseudo-code of this algorithm?
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
t;form the best possible estimate of true_value_u, by combining these
>two estimates together. In a way this is somewhat circular: we use the
>best estimate so far to compute the best new estimate. But I don't
>think that is unreasonable in t
xiefan wrote:
> could anyone help give me a link in English to buy the commercial version
>of Zen? Now I could only find some websites in Japanese, however, I could
>not read Japanese
The commercial version of Zen is not yet available for overseas people.
Thanks for your patience.
i. I guess it gives very big advantage to white
for high-dan level players. 7.0 komi would give better results.
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
scales on 19x19.
>Looking forward to desktop machines with thousands of cores
>in a few years...
What is your 23 core hardware? How much is it?
--
Yamato
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Congratulations Aja.
Do you have a plan to run AlphaGo on KGS?
It must be a 9d!
Yamato
___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
to use the name AlphaGo. So probably Aja cannot answer this, because
he does not lie.
By the way, I found this tweet interesting :)
https://twitter.com/ScienceNews/status/814559161312808965
Yamato
___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
Yes, it is AlphaGo. I am relieved that DeepMind clarified this.
Honestly I got a little frustrated that many people didn't think that
was AlphaGo. It was almost clear to me because I know the difficulty of
developing AlphaGo-like bots.
I hope Aja can comment here, also about GodMoves :)
Y
ents
since the Nature version.
Yamato
___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
uot;9D vs 9D" dataset (50,956 games).
>
> https://github.com/yenw/computer-go-dataset <https://github.com/yenw>
>
>
> ___
> Computer-go mailing list
> Computer-go@computer-go.org
&
78 matches
Mail list logo