terry mcintyre wrote:
Thank you! At present, computer go programs may be strong relative to
each other, and they may actually beat some humans of moderate
ability, especially at timescales too quick for amateur humans, but
most programs also have high-kyu-sized gaps in their knowledge,
With MCTS algorithms the error margin is high at the start of the game,
and low in the endgame. In a handicap game against a stronger opponent...
...
I did a bunch of experiments and ALWAYS got a reduced wins when I faked
the komi. But there are a million ways to do this and I may not have
My first impression of watching the game was that Leela was handicapped
by having a handicap. By that I mean it would have seen itself so far
ahead for the first few moves that is was playing arbitrarily.
In fact, Leela thought itself ahead at 80% for most of the game. It's only
in the last
Uct also has the advantage that it is much easier to use with multiple
CPUs. I know parallel alpha-beta exists, but my evaluation function is
not designed to be thread safe. If I put a big lock around it, there will
be almost no SMP scaling, since almost all the time is in the evaluation,
On Mon, 2008-08-11 at 20:39 -0700, David Fotland wrote:
Uct also has the advantage that it is much easier to use with multiple
CPUs. I know parallel alpha-beta exists, but my evaluation function is
not designed to be thread safe. If I put a big lock around it, there
will be almost no SMP
Mr. Okasaki, a strong amatur, tested MoGo with a 9 stones handicap
game at winning rate around 50% by adjusting komi on each move and
reported it played clearly stronger than others, say, on KGS and the
cluster version at Paris.
Unfortunately it sounds rather like a subjective measurement.
Every now and then, I have been known to eke out a win by running my opponent
out of time. KGS thinks that such a win is as good as any other. I am unable to
convince myself; but whenever I have a territorial advantage, I never have
second thoughts.
Terry McIntyre [EMAIL PROTECTED]
I think most programs developed by people who did not write old scool
programs has serious problems with seki. Valkyria detects some basic
seki shapes, but has problems with nakade/seki.
-Magnus
Quoting Erik van der Werf [EMAIL PROTECTED]:
You're right, my reply was sloppy (it seems I'm
On Tue, 2008-08-12 at 08:43 +0200, Gian-Carlo Pascutto wrote:
I don't like opening books. They are a liability when the rest of the
program is still improving so quickly.
I had one that worked effectively, but had to be redone if the program
improved substantially, so it was a program. I
On Tue, 2008-08-12 at 09:15 +0200, Gian-Carlo Pascutto wrote:
On Mon, 2008-08-11 at 20:39 -0700, David Fotland wrote:
Uct also has the advantage that it is much easier to use with multiple
CPUs. I know parallel alpha-beta exists, but my evaluation function is
not designed to be thread
On Tue, 2008-08-12 at 09:15 +0200, Gian-Carlo Pascutto wrote:
Aside from that, it's not theorethically necessary for alpha-beta to do
wasted work (although it will in practise), and more CPUs can make the
program worse on any practical architecture (mostly due to locking and
memory
On 12-aug-08, at 10:40, Gian-Carlo Pascutto wrote:
On Tue, 2008-08-12 at 09:15 +0200, Gian-Carlo Pascutto wrote:
Aside from that, it's not theorethically necessary for alpha-beta
to do
wasted work (although it will in practise), and more CPUs can
make the
program worse on any practical
On 12-aug-08, at 10:40, Gian-Carlo Pascutto wrote:
Well... no. Because if you have a perfectly ordered tree, in theory,
you don't need to search at all.
You need to search it to *prove* that it's perfectly ordered :-)
--
GCP
___
computer-go
On Tue, 2008-08-12 at 15:40 +0200, Gian-Carlo Pascutto wrote:
On Tue, 2008-08-12 at 09:15 +0200, Gian-Carlo Pascutto wrote:
Aside from that, it's not theorethically necessary for alpha-beta to do
wasted work (although it will in practise), and more CPUs can make the
program worse on any
On Tue, 2008-08-12 at 15:40 +0200, Gian-Carlo Pascutto wrote:
Even in the theorethical case of a perfectly ordered game tree?
I'll have to check my facts, but I remember seeing actual numbers on
this. It has something to do with the minimial tree and it was a proof
think that alpha-beta
On Tue, 2008-08-12 at 15:40 +0200, Gian-Carlo Pascutto wrote:
On Tue, 2008-08-12 at 09:15 +0200, Gian-Carlo Pascutto wrote:
Aside from that, it's not theorethically necessary for alpha-beta to do
wasted work (although it will in practise), and more CPUs can make the
program worse on any
On Tue, 2008-08-12 at 08:09 -0600, Markus Enzenberger wrote:
I don't know if you've noticed it, but the anchor Gnugo-3.7.10-a2 on
CGOS 19x19 seems to run without the correct super-ko settings. For
example, it lost game 1758 due to an illegal move.
I'm forwarding this to the group - I don't
We need to define terms so we don't end up arguing about something we
probably agree on.
Here is my assertion (which I admit needs to be checked):
Given perfect move ordering, but not a-priori knowledge of this, a
parallel program will search more nodes on average than a serial
program. And
I wrote the evaluation in the early 1980s. Multicore and threads was far
from a consideration. The big issue was how to fit all the core data in 400
KB and make it fast enough to run well on an x286 processor at about 20 MHz.
:(
I wrote the playout code in April.
David
This doesn't really
On Aug 12, 2008, at 10:44 AM, Don Dailey [EMAIL PROTECTED] wrote:
We need to define terms so we don't end up arguing about something we
probably agree on.
Here is my assertion (which I admit needs to be checked):
Given perfect move ordering, but not a-priori knowledge of this, a
parallel
From the wikipedia:
Implementations of the MTD(f) algorithm have been proven to be more
efficient (search fewer nodes) than other search algorithms (e.g. Negascout) in
games such as chess[1].
However, in practice, MTD(f) has some issues such as heavy dependence
on the transposition table,
I'm running the *-a2 anchors with the settings that Don gave me. I'll
stop them now.
Don, please send me an updated configuration file.
Álvaro.
On Tue, Aug 12, 2008 at 10:31 AM, Don Dailey [EMAIL PROTECTED] wrote:
On Tue, 2008-08-12 at 08:09 -0600, Markus Enzenberger wrote:
I don't know if
Don Dailey wrote:
We need to define terms so we don't end up arguing about something we
probably agree on.
Here is my assertion (which I admit needs to be checked):
Given perfect move ordering, but not a-priori knowledge of this, a
parallel program will search more nodes on average than a
I
did not realize that his program, even with a large tree, would not be
able to recognize the seki. I knew of course that the original Mogo
playouts had this problem, but I thought all strong programs had
solved it by now...
Hello Erik,
seki in playouts is still an unsolved problem for
There are times when self-atari is the correct play. Seki and nakade are
probably going to require a global analysis to handle properly.
Terry McIntyre [EMAIL PROTECTED]
“Wherever is found what is called a paternal government, there is found state
education. It has been discovered that the
On Aug 12, 2008, at 11:49 AM, terry mcintyre [EMAIL PROTECTED]
wrote:
From the wikipedia:
Implementations of the MTD(f) algorithm have been proven to be more
efficient (search fewer nodes) than other search algorithms (e.g.
Negascout) in games such as chess[1].
However, in practice,
Martin Mueller wrote:
I
did not realize that his program, even with a large tree, would not be
able to recognize the seki. I knew of course that the original Mogo
playouts had this problem, but I thought all strong programs had
solved it by now...
Hello Erik,
seki in playouts is still an
On Tue, Aug 12, 2008 at 6:18 PM, Martin Mueller [EMAIL PROTECTED] wrote:
I
did not realize that his program, even with a large tree, would not be
able to recognize the seki. I knew of course that the original Mogo
playouts had this problem, but I thought all strong programs had
solved it by
Jason House wrote:
Maybe the best method is to mix the top down
style of MTD(f) to drive localized alpha beta searches.
MTD(f) *is* a sequence of alpha-beta searches.
I definitely don't have all the answers.
MTD(f) doesn't parallelize any better than normal alpha-beta. The only
Gian-Carlo Pascutto wrote:
The *paper* about MTD(f) is extremely interesting because it shows
that many best-first algorithms can be rewritten as depth-first
algorithms.
It happened for SSS, it happened for proof-number search.
Who will make it happen for UCT?
Actually, there was a
On Aug 12, 2008, at 5:25 AM, Don Dailey wrote:
On Tue, 2008-08-12 at 08:43 +0200, Gian-Carlo Pascutto wrote:
I don't like opening books. They are a liability when the rest of the
program is still improving so quickly.
I had one that worked effectively, but had to be redone if the program
My fault! Don sent me a configuration file as text in email, with a
line broken down into two. He did mention it, but I was running with a
broken configuration file. It should be fine now
[EMAIL PROTECTED]:~$ pstree -ap alvaro
sshd,7704
└─bash,7705
├─cgosGtp-linux-x,7767 -c
Apparently that paper is available in Japanese:
Haruhiro Yoshimoto, Akihiro Kishimoto, Tomoyuki Kaneko, and Kazuki
Yoshizoe. Depth-First UCT and Its Application to Go (in Japanese)
(DFUCT‚̈͌é‚ւ̉ž—p‚ɂ‚¢‚Ä),
12th Game Programming Workshop in Japan, pages 30-35, 2007.
Don Dailey wrote:
On Tue, 2008-08-12 at 08:09 -0600, Markus Enzenberger wrote:
I don't know if you've noticed it, but the anchor Gnugo-3.7.10-a2 on
CGOS 19x19 seems to run without the correct super-ko settings. For
example, it lost game 1758 due to an illegal move.
I'm forwarding this to
what happens when the opponent deviates from joseki?
knowing how to punish joseki mistakes can be very,
very tricky.
also knowing which joseki to use where is very, very
sophisticated. the wrong joseki can be worse globally
than a non-joseki move.
s.
On 8/12/08, Ian Osgood [EMAIL PROTECTED]
Registration is now open for this Sunday's bot tournament on KGS. The
Formal division will be a 12-round Swiss, 9x9 boards, 8 minutes each
sudden death. The Open division will be an 8-round Swiss, 13x13 boards,
13 minutes each sudden death. They will start at 08:00 UTC (=GMT) and
08:05
On Tue, 2008-08-12 at 08:49 -0700, terry mcintyre wrote:
From the wikipedia:
Implementations of the MTD(f) algorithm have been proven to be more
efficient (search fewer nodes) than other search algorithms (e.g. Negascout)
in games such as chess[1].
However, in practice, MTD(f) has some
Just in case anyone hadn't seen the correction yet...
*
CORRECTION: *The EJ misquoted David Doshay in our 8/7 report on Computer
Beats Pro At U.S. Go Congress. What I said is that computer programs have
improved 7 to 9 stones in the last few years, [not We've improved nine
stones in just a year
On Aug 12, 2008, at 11:18 AM, steve uurtamo wrote:
On 8/12/08, Ian Osgood [EMAIL PROTECTED] wrote:
On Aug 12, 2008, at 5:25 AM, Don Dailey wrote:
On Tue, 2008-08-12 at 08:43 +0200, Gian-Carlo Pascutto wrote:
I don't like opening books. They are a liability when the rest
of the
From: steve uurtamo [EMAIL PROTECTED]
what happens when the opponent deviates from joseki?
knowing how to punish joseki mistakes can be very, very tricky.
From my observations at the mogo-vs-pro game, given lots of time and CPU
cores, Mogo is able to discover how to punish such deviations. In
On Tue, 2008-08-12 at 10:41 -0700, Ian Osgood wrote:
On Aug 12, 2008, at 5:25 AM, Don Dailey wrote:
On Tue, 2008-08-12 at 08:43 +0200, Gian-Carlo Pascutto wrote:
I don't like opening books. They are a liability when the rest of the
program is still improving so quickly.
I had one
You can also get rid of -score aftermath, I am told it does nothing
when gtp mode is active.
- Don
On Tue, 2008-08-12 at 13:43 -0400, Álvaro Begué wrote:
My fault! Don sent me a configuration file as text in email, with a
line broken down into two. He did mention it, but I was running with a
On Tue, 2008-08-12 at 18:18 +0200, Gian-Carlo Pascutto wrote:
Don Dailey wrote:
We need to define terms so we don't end up arguing about something we
probably agree on.
Here is my assertion (which I admit needs to be checked):
Given perfect move ordering, but not a-priori
I had asked Chris to print the correction, and he was glad to do so.
Now I am trying to get the article on the AGA web site updated. I have
already seen a bunch of things on the internet that link to it and the
mistaken comment (9 stones in a year, computer superiority real soon)
is
Don Dailey wrote:
Here is an important snippet, but proofs follow in the paper:
The critical path length C is the time it would take for the program
to run on an infinite-processor machine with no scheduling overheads.
Note that it doesn't mention anything about useful WORK, because this is
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
Well you can blame me for linking to the AGA story on Slashdot, but at
least I didn't repeat the misquote, and I also asked Chris to fix it
on the AGA site. I figured it would probably make Slashdot quickly
anyway, so the story might as well be written by someone with at least
a bit of a
Hi Bob,
No problem at all ... it was inevitable that this would happen. I just
found that the Oregonian, which printed that I wrote GNU Go, properly
changed it to SlugGo in the on line version.
What I am trying to understand now is how the New Scientist technology
blog has me properly
Rémi Coulom: [EMAIL PROTECTED]:
Gian-Carlo Pascutto wrote:
The *paper* about MTD(f) is extremely interesting because it shows
that many best-first algorithms can be rewritten as depth-first
algorithms.
It happened for SSS, it happened for proof-number search.
Who will make it happen for
Hideki Kato: [EMAIL PROTECTED]:
Rémi Coulom: [EMAIL PROTECTED]:
Gian-Carlo Pascutto wrote:
The *paper* about MTD(f) is extremely interesting because it shows
that many best-first algorithms can be rewritten as depth-first
algorithms.
It happened for SSS, it happened for proof-number
On Wed, Aug 13, 2008 at 1:29 PM, Hideki Kato [EMAIL PROTECTED] wrote:
Rémi Coulom: [EMAIL PROTECTED]:
Gian-Carlo Pascutto wrote:
The *paper* about MTD(f) is extremely interesting because it shows
that many best-first algorithms can be rewritten as depth-first
algorithms.
It happened for
Interesting. Could you (or someone else) explain how DFUCT works? I'd
imagine it doesn't save all the nodes in memory, but that seems rather
counterintuitive.
___
computer-go mailing list
computer-go@computer-go.org
All,
Can anyone detail the design of the version of Mogo that beat the
professional? Or is there a web-page where at least the general approach
has been described? Is the information even public? I am not seeing the
the implementation details, just the overall design and general
strategies.
-Original Message-
From: [EMAIL PROTECTED] [mailto:computer-go-
[EMAIL PROTECTED] On Behalf Of Ian Osgood
This is a different kind of opening book than I'm thinking of. You
are both talking about cached computation, whereas I consider an
opening book as codified theory and wisdom
Can anyone detail the design of the version of Mogo
You can find some papers here:
http://www.lri.fr/~gelly/Publications.htm
I don't personally know of any papers detailing improvements or changes
in the past year (I think they've mainly been on making it more parallel).
Darren
--
FYI the bayes rated games updates every 6 hours. The three different
boards sizes are staggered so that every 2 hours one of them updates.
- Don
On Mon, 2008-08-11 at 17:22 -0400, Don Dailey wrote:
The main page now has links to a bayes rated chart for each board
size.
This will be
How long does bayes rankings take to run? If I understand the math,
you should be able to feed the prior output back in for a dramatic
speed gain.
I'd really like to see bayes rankings with the same display filter as
the normal standings.
Sent from my iPhone
On Aug 12, 2008, at 10:53
I hope that David Fotland can chime in here on value of joseki
libraries on program strength.
Also, which existing classical program is considered the best semeai
player?
Ian
I don't know that joseki knowledge mad Many Faces stronger. Go Intellect
always used to turn off the joseki
Zach Wegner: [EMAIL PROTECTED]:
Interesting. Could you (or someone else) explain how DFUCT works? I'd
imagine it doesn't save all the nodes in memory, but that seems rather
counterintuitive.
As far as I understand,
The basic idea is that DFUCT continues searching avoiding going back
to root
Oops.
Hideki Kato: [EMAIL PROTECTED]:
Zach Wegner: [EMAIL PROTECTED]:
Interesting. Could you (or someone else) explain how DFUCT works? I'd
imagine it doesn't save all the nodes in memory, but that seems rather
counterintuitive.
As far as I understand,
The basic idea is that DFUCT continues
The MoGo programmer who answered questions after the match (Olivier
Teytaud) did state that MoGo no longer used UCT. He gave a one-line
statement of the reason they switched, which I did not follow.
Bob
On Aug 12, 2008, at 7:00 PM, Jim O'Flaherty wrote:
All,
Can anyone detail the design
Bob Hearn: [EMAIL PROTECTED]:
The MoGo programmer who answered questions after the match (Olivier
Teytaud) did state that MoGo no longer used UCT. He gave a one-line
statement of the reason they switched, which I did not follow.
The first post Olivier wrote they no longer used UCT is:
The MoGo programmer who answered questions after the match (Olivier
Teytaud) did state that MoGo no longer used UCT. He gave a one-line
statement of the reason they switched, which I did not follow.
For me, and I suspect some others here, UCT has been used to mean the
way of expanding out the
63 matches
Mail list logo
