I am not good at term definition, but I would say RAVE is the algorithm
extending AMAF in MCTS, including how to accumulate the counts at each node
(trivial extension even though implementation can be tricky), how to
combine with UCT (or other move choice), and how to integrate with priors
(based
Hi,
Olivier answered for the new version.
On the downloadable version, I don't remember exactly (almost 2 years
back now...), but I think Mogo will still pass if all the other moves
are clearly loosing. So it should understand somehow Seki
situations.
If that is correct, the sentence is not
Obviously I should read better the emails before answering. Olivier
rightly answered for all versions.
Sorry,
Sylvain
On Tue, Jun 30, 2009 at 7:59 PM, Sylvain Gellysylvain.ge...@m4x.org wrote:
Hi,
Olivier answered for the new version.
On the downloadable version, I don't remember exactly
On Tue, Jun 30, 2009 at 12:47 AM, Peter Drakedr...@lclark.edu wrote:
A while back, Sylvain Gelly posted this code:
ChooseMove(node, board) {
bias = 0.015 // I put a random number here, to be tuned
b = bias * bias / 0.25
best_value = -1
best_move = PASSMOVE
for (move in board.allmoves
2009/5/1 Brian Sheppard sheppar...@aol.com:
In reading Sylvain Gelly's thesis, it seemed that incorporating a prior
estimate of winning percentage is
very important to the practical strength of Mogo.
E.g., with 1 trials, Mogo achieved 2110 rating on CGOS, whereas my
program attempts to
On Wed, Apr 8, 2009 at 10:46 AM, Darren Cook dar...@dcook.org wrote:
End game is another issue. MC programs only aim on winning, so they
endgame is nor perfect in sense human would define it, but perfect
enough to win if the game is winnable.
You can modify komi to get the human expert and MC
Hi Mattew,
I cannot answer for the current version of Mogo but I can for the one
1.5 years ago. Maybe it still holds.
We had a transposition table as it was designed like that from the
beginning. However the prior value of the node, was initialized when
the node was created and indeed was
On Fri, Jan 30, 2009 at 9:29 AM, Magnus Persson magnus.pers...@phmp.sewrote:
Quoting Sylvain Gelly sylvain.ge...@m4x.org:
On Wed, Jan 28, 2009 at 10:01 PM, Isaac Deutsch i...@gmx.ch wrote:
And a final question: You calculate the (beta) coefficient as
c = rc / (rc+c+rc*c*BIAS);
which
On Wed, Jan 28, 2009 at 10:01 PM, Isaac Deutsch i...@gmx.ch wrote:
Hi again ;)
I found some time to actually implement this stuff. And, this has raised
some small questions. In this part of the code:
for (j = index; j moves_played_in_tree.size(); j += 2) {
//stuff
}
for (j =
Hi,
I did not mention here the prior initialization that is done in each node.
When you create a node, you can look at all possible move and if a pattern
matches (the exact same as in the playout) you initialize rw and rc to 14.
If the move saves a capture (same as in the playout), same
2009/1/21 Olivier Teytaud olivier.teyt...@lri.fr
Of course you can do put much more clever prior if you are a player and
know the subtleties of the game.
E.g. patterns extracted from databases - but it's not enough, carefully
tune the coefficients for empty triangles (important!) and
Hi,
You should really look into never deallocate memory (by calling delete/free)
but keeping it in some memory pool. I did that for the main objects that you
deal with: nodes and small vectors (the one you create on the fly to keep
the moves that have been played in the playout). It really speeds
Hi,
Sorry for the slow reply.
After those discussions, I figured out that pseudo code was the
fastest clear and not ambiguous way to describe how to precisely
implement the algorithm. I needed to find some time to write it.
Note that I did not write only the backup phase because to clearly
, already_played,
moves_played_out_tree)
Backup(black_wins, nodes_seen_in_tree, moves_played_in_tree,
moves_played_out_tree, already_played)
}
2009/1/17 Sylvain Gelly sylvain.ge...@m4x.org:
Hi,
Sorry for the slow reply.
After those
strength. If there's any noticeable
difference I'll try to modify the code in my reference implementation to
reflect the 'correct' method.
Mark
On Jan 17, 2009, at 11:48 AM, Sylvain Gelly wrote:
Hi,
Sorry for the slow reply.
After those discussions, I figured out that pseudo code
Hi Issac,
You are welcome, and I am happy there is finally a clearer of implementing
RAVE out there. I believe I should have done it much earlier, sorry for
that, but better late than never, no? :)
Best,
Sylvain
2009/1/17 Isaac Deutsch i...@gmx.ch
Hi,
Sorry for the slow reply.
Hi
I'd
Good catch :)Indeed it makes no sense with the *, sorry...
Sylvain
2009/1/17 Magnus Persson magnus.pers...@phmp.se
I think I found a bug in ChooseMove
Quoting Sylvain Gelly sylvain.ge...@m4x.org:
coefficient = 1 - rc * (rc + c + rc * c * b)
I think this has to be
coefficient = 1
2009/1/10 Isaac Deutsch i...@gmx.ch
Hi Sylvain,
I think it's starting to make sense now. :-)
Sorry to be unclear. I wish we have a white board where we could discuss
and
that would sorted out in a few minutes :).
Several results turn up in a google search ;p
Hi Isaac,
in a nutshell RAVE is basically AMAF adapted for Monte Carlo Tree Search.
The original paper describing it is
http://www.machinelearning.org/proceedings/icml2007/papers/387.pdf and a
paper for broader audience can be found here:
http://www.lri.fr/~gelly/paper/MoGoNectar.pdf (the picture
Hi Isaac,
2009/1/9 Isaac Deutsch i...@gmx.ch
Hi Sylvain,
Thanks for your quick answer.
in a nutshell RAVE is basically AMAF adapted for Monte Carlo Tree Search.
The original paper describing it is
http://www.machinelearning.org/proceedings/icml2007/papers/387.pdf and a
paper for
What I did (was a long time ago, I don't know if it is still used in
Mogo), is to compute the m best moves every so often and most of the
time just do the max over those m moves.
m was on the order of 5, and every so often was an increasing
function like sqrt or log (I don't remember).
That speeds
C++ on linux (with a port on windows using cygwin libraries for the binary
release)
Sylvain
2008/8/13 steve uurtamo [EMAIL PROTECTED]
And what language/platform is Mogo written in; C/C++, Java, Assembly,
PHP,
etc.?
This made coffee spray out of my nose (PHP).
I think that C is most
the mistaken comment (9 stones in a year, computer superiority real soon)
is getting repeated a huge number of times.
As one of my computer science teacher said: if your editor has the
copy/paste feature, throw it away.
It obviously applies to programming and apparently to publication as well
2008/7/28 Ray Tayek [EMAIL PROTECTED]
At 07:53 PM 7/27/2008, you wrote:
The traditional programs are around 10 kyu, but the new ones are 2 to 4
kyu,
at least on KGS. I've seen some handicap games against dan players that
are
consistent with these ratings.
wow. that's impressive. can
You can download for free an old version of MoGo (which reached 2k on KGS
on a 4 CPU machine) at:
http://www.lri.fr/~gelly/MoGo_Download.htmhttp://www.lri.fr/%7Egelly/MoGo_Download.htm
http://www.lri.fr/~gelly/MoGo_Download.htmhttp://www.lri.fr/%7Egelly/MoGo_Download.htm
the exe just sits
MoGo has a notion of internal node in the tree (as most of the UCT
programs I think) and the state-action pairs are only kept for those.
Sylvain
2008/7/2 Jason Galbraith [EMAIL PROTECTED]:
I've been looking at RAVE (Rapid Action Value Estimate), which MoGo uses. The
score of states during
This seems to be the case and I still do not really on some level
understand why. Since with the chinese go rules the board should be
effectively stateless (exempting ko information) all the information be
contained in the the current position. Why additional information is needed
i.e. an
It was 2 cores 2.6GHz. (intel core2 duo).
2008/3/21, Olivier Teytaud [EMAIL PROTECTED]:
What computing power did have that MoGo at its disposal?
4 cores, 2.4 GHz.
___
computer-go mailing list
computer-go@computer-go.org
So MoGo may be
using a floating point function to estimate the score of a playout,
otherwise there would be no reason to use floating point. But I may be
guessing wrong. Maybe they can tell us ?
We don't (at least up to the release, I don't know everything they are
doing now). Using the
2008/2/22, Alain Baeckeroot [EMAIL PROTECTED]:
Le jeudi 21 février 2008, Don Dailey a écrit :
If you look at the table you will notice that going from level 4 to
level 11 (which is 7 doublings and should take 128X longer) only takes
59.43 X longer.
So if we plot 9X9 rank vs time,
Hi Don,
2008/2/21, Don Dailey [EMAIL PROTECTED]:
If you look at the table you will notice that going from level 4 to level
11 (which is 7 doublings and should take 128X longer) only takes 59.43 X
longer.
Mogo's stop early heuristic works better at longer levels.
That is actually very
Hi Hideki,
Isn't possible to just redirect stderr?
Best,
Sylvain
2008/2/14, Hideki Kato [EMAIL PROTECTED]:
Hi Olivier,
New versions of MoGo don't put log files, which was very useful to
study. Don't you have plan to release such versions put log files?
-Hideki
Olivier Teytaud:
Thinking a little more about it, I think we have to add an hypothesis
which is that, for a given move, the number of AMAF updates if alpha
(nb total UCT updates), with alpha 1. That seems to hold for most of
the updates (with alpha close to 0.5), but there may be cases where it
does not
As far as I see,
if RAVE gives constant value 0 to one move, it will never be tested if
other moves
have non-zero AMAF values.
A move
with real empirical probability 0 of winning and AMAF value of 0.01
will always be preferred to a non-simulated move with AMAF 0.0, whatever
may be
the
Hi all,
I added those downloads on the MoGo's download page:
http://www.lri.fr/~gelly/MoGo_Download.htm
Cheers,
Sylvain
2008/2/9, Olivier Teytaud [EMAIL PROTECTED]:
For people requesting mogoRelease3 without the bug for long computation
times due to a float instead of a double:
A new position is always visited unless the leaf of the tree is the
end of the game. In that case, one player always win, so the other
always win. Then, the losing player will explore all the other moves
to avoid the sure loss. If all moves are still loosing, that will
propagate to the move
Hi Erik,
(1) They compared Rave to plain UCT. If they would have compared it to
a more sophisticated implementation (like the best Mogo before Rave)
they probably could not have shown a spectacular improvement.
The best Mogo before Rave was very close to plain UCT with the
sequence-like
Hi Erik,
In the ICML version of UCT without RAVE, you did not use your First
Play Urgency, right?
I think that using FPU has an effect similar to what others reported
with their progressive widening. From what I've seen it looks like
plain UCT, without FPU or progressive widening, has more
No problem for me. I did not want to multiply the number of versions not to
confuse people. With the double version, don't forget it will increase the
memory footprint for a given number of nodes.
Sylvain
2008/1/30, Olivier Teytaud [EMAIL PROTECTED]:
I can provide a new release with double
Hi,
With such a large number of playouts, the tree size limit (and so
heavy pruning) is certainly a possible hypothesis. The simplest way to
test it would be to run the same MoGo_17 or _18 with a much bigger
tree (taking more memory). --collectorLimitTreeSize is by default
40 (number of
to this situation?
Terry McIntyre [EMAIL PROTECTED]
- Original Message
From: Sylvain Gelly [EMAIL PROTECTED]
To: computer-go computer-go@computer-go.org
Sent: Tuesday, January 29, 2008 10:36:38 AM
Subject: Re: [computer-go] 19x19 Study
but not linearly and you can see a nice gradual curve
Google finds it:
http://tao.lri.fr/Papers/thesesTAO/SylvainGellyThesis.pdf
That is NOT the latest version. Please at least let me put the latest
version on my web site, it took me so long to correct it :).
Sylvain
___
computer-go mailing list
The reason I said that was this behavior from mogo. If I start it without
that switch and as for a move, it allocates 20 seconds. If I then issue a
small
time_left command and ask for another move, it allocates a much smaller
amount of time. Here is the output:
Because you give a
and MoGo expects a
gtp command to override it.
Cheers,
Sylvain
2008/1/13, Michael Williams [EMAIL PROTECTED]:
Sylvain Gelly wrote:
The reason I said that was this behavior from mogo. If I start it
without that switch and as for a move, it allocates 20 seconds. If
I then issue
It looks like MoGo does respect the time_left commands from GTP, so I
don't think the totalTime parameter is required in this case.
What do you mean? If you don't put --totalTime, then MoGo indeed ignores
time_left. If you put --totalTime, then it respect the time_left.
Cheers,
Sylvain
2008/1/10, [EMAIL PROTECTED] [EMAIL PROTECTED]:
Hi Sylvain,
Have you finished your thesis? We are eager to read it:-)
Hi,
Yes I did! :).It is not on my website, but will (soon?).
However, you should not be so eager to read it :)
Cheers,
Sylvain
On 1/10/08, Sylvain Gelly [EMAIL PROTECTED
Hi,
I guess the public version of MoGo was designed with a
focus on 9x9 and not 19x19.
It was not more on 9x9 that 19x19, it was more or less the best settings of
MoGo against gnugo at the moment I left the developpement (early september)
for both 9x9 and 19x19.
Or is there something else I
You should be using area scoring only and if you are playing handicap
games then either YOU or MOGO is not counting them the same. Or
perhaps Mogo has a bug in the handicap code.
MoGo uses KGS handicap counting (add 1 point to white for each handicap
stone) if the GTP set_handicap_stones
Hi,
2007/10/8, Benjamin Teuber [EMAIL PROTECTED]:
Hi everybody - especially Sylvain =)
I'm wondering whether the formula to determine the balance between RAVE
and UCT,
beta = sqrt(c / 3 * parentVisits + c),
has any mathematical background - or is it just a best guess for something
that
Hi,
Yes you can:
--nbTotalSimulations 3000
Once you set this option it ignores all other time settings.
Cheers,
Sylvain
2007/10/7, Yamato [EMAIL PROTECTED]:
Sylvain,
Can I set the number of the simulations per move instead of the
thinking time? (like --simulations 3000)
If possible, it
that means that MoGo actually uses twice as much time, ie 60
s.
Sorry for the confusion for you, and for other people who may have run into
it.
Cheers,
Sylvain
2007/10/6, [EMAIL PROTECTED] [EMAIL PROTECTED]:
I used --19 --time 30 --nbThreads 2
DL
-Original Message-
From: Sylvain Gelly
I set 30 second playing
time, but often it takes up to couple minutes for a move.
That is not normal, could you post the command line you use?
Cheers,
Sylvain
___
computer-go mailing list
computer-go@computer-go.org
Hi Darren,
Thank you for playing (I hope you enjoyed :)) and your report.
Sylvain, can you explain more about why it has this particular weakness?
What you describe is exactly the issue I describe in the FAQ.
From my analysis this come from the fact that the simulations are more
likely to make
Hi,
Has anyone else done scaling experiments with 19x19 and UCT?
I did some months ago, and reported them in that list with the title
19x19 Go, scalability with time vs handicap
(http://www.mail-archive.com/computer-go%40computer-go.org/msg02775.html)
The results are old, but now everyone can
Hi Darren,
It preserves the tree if and only if you add:
--pondering 1
If you don't want to use pondering, but you still want to keep the tree
between moves, add
--keepTreeIfPossible 1
(not documented, and from my memory, it may be not the right option :p)
Hoping this helps,
Sylvain
Hi Gilles,
yes they are some problems to use MoGo with Drago. The main issue is the
initial message written to stderr as guessed by Dave. Actually, Drago
handles incorrectly stdout and stderr in the same way but this is easily
corrected.
Good news!
I have uploaded a patch for using MoGo
Hi Hideki,
The Windows version, however, seems much weaker than MoGo that running
on KGS these days on 19x19, even giving much longer time setting such
as --time 300 for example. I guess some other settings than time
are necessary.
Sorry, you are right the 19x19 settings always put the
Windows version being the same
strength as KGS version?
Regards,
Hideki
Sylvain Gelly:
[EMAIL PROTECTED]:
Hi all,
you can find here:
http://www.lri.fr/~gelly/MoGo_Download.htm
an update of MoGo's release, especially binary for non pentium4
compatible processors, some other options
Hi Dave and thanks,
Drago works well for Gnugo and also for my program. Mogo starts but then
exits prematurely with a message from Drago abnormal termination of
engine.
:-(
One thing, when I run it from the command line, it spits out a lot of
non-gtp format diagnostic information
Hi all,
you can find here:
http://www.lri.fr/~gelly/MoGo_Download.htm
an update of MoGo's release, especially binary for non pentium4
compatible processors, some other options explained, and maybe more
interesting, an option for time management (I stupidly did not think
that people would use
Hi Hideki,
Some computer-go friends in Japan have reported that even current
binary of MoGo doesn't work on Athlon XP or Celeron. Both (and
Pentium III) have no SSE2 instructions while Pentium 4 has.
Ok, I have to compile for older processor too then (I did not expect
so old proc were still
I had a similar issue where pthread/gcc/cygwin combination produced a very
slow application. I had better success with Visual C++ and Boost (for
portable threads).
You are right, I should have used Boost for the threads...
Unfortunately, I used pthread, and that mean that the threading part
2007/9/10, David Stafford [EMAIL PROTECTED]:
What are the options for someone who would like a dan-level opponent (even if
it's 9x9)
but doesn't have a Linux system currently? Are there choices other than
MoGo? If not,
I'm willing to build a Linux box but I have some questions:
- Is a
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.
Sylvain
___
computer-go mailing list
computer-go@computer-go.org
2007/9/10, Don Dailey [EMAIL PROTECTED]:
The command line parameter to change board size does nothing.
I tried:
./mogo --19
./mogo -19
and it only seemed to want to play on 9x9 boards.Am I doing
something wrong?
As I explained earlier in an answer, the command line parameters
I'm a little confused. If I operate with no parameters it works ok,
No parameters means --19
but if I do ./mogo --7 (for instance) it goes into some kind of
self-training mode.
Did you see a --7 option on the manual? :-p
There is no --7 option, nor a --13 one. You should put a --9 or a
Well, I'm hoping for a Mac version someday...
Hopefully it will happen. As I don't have a Mac, I rely on external
help. I'll let you know :).
Sylvain
___
computer-go mailing list
computer-go@computer-go.org
tried to open opening, success 0-- in grey
Here it does not find the file, because the file is with the binaries
and gogui (at least your version) looks into the gogui/bin directory.
But that does not prevent MoGo to work.
I guess so, as when I added
when I run the Linux exeutable on my Fedora 8/Athlon XP, I get a
coredump:
$ mogo --9 --time 12
Load opening database opening succeed (nbEntries=618) (nbIllegalMoves removed
0)
tried to open opening, success 1
Illegal instruction (core dumped)
could it be that it is compiled for specific
Is there a option like gnugo's --capture-all-dead?
In my test(./mogo --9 --time 1), seems mogo passed when not capture
alldead stones.
As this release is mainly for humans to play, it is set to play
against humans, so passing as soon as the opponent passes and it is
safe to pass.
If you
I think gogui is in fact looking for files in the directory from which
it is launched. Try this to copy the opening database in this directory.
Yes exactly, thank you Guillaume for explaining better than I can :p
It runs perfectly on an Opteron 2.6GHz.
Good!
But not on a Power5+ processor.
Hi all,
Thank you for all your comments and reports, and I am pleased some of
you are happy to use it. Please feel free to share the links,
especially for players who do not read this list.
I am sorry it does not work for some of you. I will look into it as
soon as I can.
BTW, I tried to answer
I guess the search path you've coded is something wrong or different
depends on the distributions.
The search path is simply .
I'd like to suggest to use some
environment variable dedicated to mogo.
I think the recent version of gogui let you define the working
directory. Also, as Guillaume,
the auto-number
feature, mogo worked fine.
# Settings - Configure Shell - Auto number
Cheers,
Hideki
Sylvain Gelly: [EMAIL PROTECTED]:
I guess the search path you've coded is something wrong or different
depends on the distributions.
The search path is simply .
I'd like to suggest
auto-numbering in GoGui prepends all commands with an integer ID,
which is sent to the program and should be used by the program in
its response, see the GTP specification.
Ok, I did not know that, thanks. So that part of GTP is simply not
supported in MoGo :).
Cheers,
Sylvain
Hi Markus, Hi all,
I updated the package to fix the issues you get and some other minor
ones. Please update before reporting a problem, and please report any
further problem :-).
I don't know about Ubuntu, but the default GCC configuration on Fedora
does not set CPU-specific compiler options,
Hi all,
I am pleased to announce a binary release of current version of MoGo.
It is specially designed for players but of course it may be
interesting for some of you as a benchmark.
You download it and see the instructions there:
http://www.lri.fr/~gelly/MoGo.htm
Of course, please feel free to
2007/9/9, Brian Slesinsky [EMAIL PROTECTED]:
Are there any plans to release the source?
I don't think so. Plus, some will work on MoGo source code, so it is
their decision, not mine.
Perhaps someone else will figure out how to port it.
Well, it actually builds and work on windows, only the
Try MinGW (and MSYS). MinGW has GCC ver. 4.2.1.
http://sourceforge.net/project/showfiles.php?group_id=2435
Yes, I saw that and tried. But the thing is that MoGo use pthread
library for multitreading, and, as far as I know, MinGW does not
provide pthread (does it?). It is why I needed cygwin.
Hi Chrilly,
I am sorry about your fight with a dog, and I hope you are ok!
I read your slides: interesting point of view, whereas you seem a
little frustrated. Thank you for sharing your opinion.
However, I have to disagree with this statement:
UCT: Complete Antithesis to AI-approach
I really
Hi Chrilly,
It was always the goal of McCarthy and his followers to simulate and to surpass
the human mind. (...)
UCT has nothing to do with human Go. It has some similarity to the behaviour
of ant-collonies (its not in the technical sense an ant-colony algo). It was
never the goal of AI to
Hi Chrilly,
Take a look at this list, there are already maybe more than 100 posts
on this subject. While I agree with you, just don't worry, almost all
computer go games are with the same set of rules, just ignore the
rest.
Cheers,
Sylvain
2007/7/12, chrilly [EMAIL PROTECTED]:
I am playing
Hi David,
(...) I cannot imagine that progress will be
made without a great deal of domain knowledge.
Depending on what you exactly mean I disagree.
I mean progress by the standard usually applied to computer Go:
programs that can beat 1D humans on a full board, and then get
better.
For me
Hi Chrilly,
1) there are database of thousands of professional games for few
dollards. There are not 9x9, but (i) making database is not making
progress in the field, it is just having some temporary advantage in
tournaments. (ii) Opening is much less important in Go than in Chess,
it is why we
Hi Don,
This is a very interesting study!
Sylvain
2007/6/25, Don Dailey [EMAIL PROTECTED]:
Someone just reminded me of the scalability study I did a few months
back and I reported that I would continue to run it for perhaps a few
more weeks.
I did run about 20% more games, but the data was
2007/6/23, Yamato [EMAIL PROTECTED]:
Using prior knowledge on normal uct, and this was the use of prior
knowledge brought about the same improvement.
You mean, there is more improvement when using both?
I mean that there is no need to have AMAF to get improvement by using prior
knowledge.
Hello all,
We just presented our paper describing MoGo's improvements at ICML,
and we thought we would pass on some of the feedback and corrections
we have received.
(http://www.machinelearning.org/proceedings/icml2007/papers/387.pdf)
The way that we incorporate prior knowledge in UCT can be
interesting to see what mogo thinks on those variations.
Best Regards,
Lukasz
On 6/14/07, Sylvain Gelly [EMAIL PROTECTED] wrote:
Hello Sanghyeon, thank you for your comments.
After white (mogo) H2, MoGo was estimating 74%, and expecting:
H2 G1 H3 B1 A1 B3 H1 F8 B5 H4
This is far too optimistic
Hello Sanghyeon, thank you for your comments.
After white (mogo) H2, MoGo was estimating 74%, and expecting:
H2 G1 H3 B1 A1 B3 H1 F8 B5 H4
This is far too optimistic. Why would black play H2? :-)
Sorry, white played H2. The sequence I gave starts with white move :).
Black was expecting to
Hello John,
Thank you for your interest.
Figure 3 in your UCT paper shows the accuracy of different simulation policies.
Could you repeat these experiments for accuracy of win/loss determination only?
Actually the labelled positions are rather end game positions, and are
labelled as 0/1
Hello,
When writing C/C++ for multi-platform student assignments using gcc,
we always used the args:
-ansi -Wall -pedantic
Maybe it depends on the gcc versions, but I always use -Wall -W rather
than only -Wall. -W turns on (important) warnings which are not turned
on with only -Wall, and as
Hi Rémi,
Thank you for this paper. I found the work very interesting, well
written, and the paper is clear and pleasant to read.
As two things are modified in the same time (simulation policy and
tree search), I wonder what is the contribution of each part in the
overall improvement. For example
Hi Rémi,
2007/5/17, Rémi Coulom [EMAIL PROTECTED]:
to Sylvain: Here are tests of Crazy Stone at 90s/game 1CPU against GNU
3.6 level 10, measured over about 200 games
[...]
Thank you for your answer. These numbers are interesting.
The improvement in the tree search is really huge. It is what I
Hello Daniel,
With the addition of fuseki and joseki library will its rating increase?
Especially a fuseki library. Will a fuseki library be consistent with its
playing style?
That is an interesting and not trivial question. The problem is that the
player has somewhere to understand the
2007/4/11, [EMAIL PROTECTED] [EMAIL PROTECTED]:
I watched MoGo playing with different rank of players. Usually 5d players
has no problem winning. Starting from 4d begin to lose games. However, part
of it is due to most players are not familar with 9x9 Go. Taking this into
consideration I place
Hello,
I'm curious to know, how many playouts (in Sensei's 100k
is mentioned for CGOS) MoGoBot plays, i.e., how serious
version is it?
This version plays on a intel core2 duo, and on a 10 minutes game, it makes
between 40 and 5 playouts per move (more at the beginning). The
current
I also find this kind of information very interesting and useful. Now I have
a better feel for what kind of scaling is realistic to try for and how to
measure it.
Putting some recent data points together, it look like giving Mogo 2 orders
of magnitude more computer power would result in low dan
Hello,
2007/4/6, Tom Cooper [EMAIL PROTECTED]:
My guess is that the complexity of achieving a fixed standard of play
(eg 1 dan) using a global alpha-beta or MC search is an exponential
function of the board size.
(...)
To some extent, this is testable today by finding how a global search
Here's another way to test this sort of thing that is completely
intrinsic to the engine (doesn't require gnugo):
Start with and empty board and zero komi. Analyze using UCT until the
winning percentage at the root reaches X. Note the number of
simulations required (or the amount of time).
I can turn the difficulty settings way down so that I have a chance to
actually win a game or two.
You can always decrease the time per move and at some limit, you'll reach
random play. Even if I can't win against MoGo with 300 playouts per move (I
am so bad :-( ), but can I beat a random
1 - 100 of 172 matches
Mail list logo