On Aug 9, 2008, at 11:46 AM, Don Dailey <[EMAIL PROTECTED]> wrote:
On Sat, 2008-08-09 at 10:42 -0400, Jason House wrote:
I hit this problem long ago when CGOS was young. The fix at that time
was to send the estimated time until the next round. Eventually, that
cluttered the logs a
I like the idea of using Bayesian ELO ratings instead. They should
adapt better and faster. It would give better rank confidence than the
current k factor. For example, kartofel would have kept a low
confidence.
Sent from my iPhone
On Aug 10, 2008, at 11:51 AM, "David Fotland" <[EMAIL PROT
On Aug 11, 2008, at 12:02 PM, "Erik van der Werf" <[EMAIL PROTECTED]
> wrote:
On Mon, Aug 11, 2008 at 4:54 PM, Gian-Carlo Pascutto <[EMAIL PROTECTED]>
wrote:
Some time ago I observed that kgsgtp does not tell my program that the
opponent has resigned (which is a bit annoying because it the
On Aug 11, 2008, at 2:06 PM, Don Dailey <[EMAIL PROTECTED]> wrote:
On Mon, 2008-08-11 at 17:26 +0200, Rémi Coulom wrote:
Basti Weidemyr wrote:
What would you have done in a case like this? :)
You could not declare that game a win for the computer and survive.
Yes, and I really hate this.
Sent from my iPhone
On Aug 11, 2008, at 4:00 PM, Don Dailey <[EMAIL PROTECTED]> wrote:
I would be angry if I worked hard to control my time usage, only for
my
opponent to be forgiven at my expense, despite the rules.
Hmmm... This sounds very familiar...
_
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 pr
tyranny in the nursery.”
Benjamin Disraeli, Speech in the House of Commons [June 15, 1874]
- Original Message
From: Jason House <[EMAIL PROTECTED]>
To: "[EMAIL PROTECTED]" <[EMAIL PROTECTED]>; computer-go >
Sent: Tuesday, August 12, 2008 8:17:35 AM
Subject: Re: [comput
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 PM,
If this is aimed at clearing up ambiguity, you should state which way
the handicap was given.
Sent from my iPhone
On Aug 13, 2008, at 2:08 PM, "Chaslot G (MICC)" <[EMAIL PROTECTED]
> wrote:
Dear all,
There were details that were unclear about the victory of MoGo.
Hence I created a website
I think it's be generally useful. Someone else recently posted about
post-game pondering problems.
I can commit this code this weekend, but can not post a release. I
keep hoping Don will embrace file releases through SourceForge since
we both have access to it...
Sent from my iPhone
On A
On Aug 18, 2008, at 10:19 AM, Don Dailey <[EMAIL PROTECTED]> wrote:
Perhaps on the crosstable page I should put another column that tells
you, for each opponent, what your winning expectation is based on your
respective ELO ratings? It would be nice to see at a glance whether
you are winning m
One small correction, my server wasn't running two copies of HouseBot,
it was running three. I discovered later there was a 4th running from
somewhere else but haven't tracked it down yet.
Sent from my iPhone
On Aug 18, 2008, at 12:08 PM, Nick Wedd <[EMAIL PROTECTED]> wrote:
My report is a
HouseBot's UCT unit tests build trees to test search time and search
results. This is done by using a simplified (non-go) game where I can
define the game state and win rate at different points in the tree.
Sent from my iPhone
On Aug 29, 2008, at 5:38 AM, "Ingo Althöfer" <[EMAIL PROTECTED]>
Sent from my iPhone
On Sep 4, 2008, at 1:49 PM, Gian-Carlo Pascutto <[EMAIL PROTECTED]> wrote:
As far as I understand, GTP only supports canadian byo-yomi which is
not
so popular. Leela should support it but I have never tried.
kgs-time_settings is much easier to interpret. KGS will send
Actually your summary of what people do sounds exactly like what MC
programs do, except for one point...
MC programs don't differentiate moves by point value. They only look
at winning rate. It's extremely tough to differentiate the one move
sequence with 99.1% win rate when all other moves
On Sep 8, 2008, at 11:45 AM, "Olivier Teytaud"
<[EMAIL PROTECTED]> wrote:
By my recent experiments, 8~9 * (threads - 1) ELO is lost. This
matches my earlier result well.
Do you have tricks for avoiding redundancies between simulations ?
I suggest simple tricks like "do not go to node X if
Don,
Once upon a time, you described the heuristics used in the various 9x9
bots you run on CGOS. I tried googling for the old message on computer
go, but came up empty handed.
If I remember correctly, it does AMAF with a few biased in move
selection/playouts. I think it looked at the pro
HBotSVN's processor details are empty, and there seems to be confusion about
the end of the round 6 game. I hope the additional detail below is helpful.
Processor:
- One core of an "Intel(R) Core(TM)2 Duo CPU T5450 @ 1.66GHz"
Open Division, Round 6:
- I noticed that KGS was non-responsi
On Sep 22, 2008, at 7:59 AM, Magnus Persson <[EMAIL PROTECTED]>
wrote:
In the case of the ladders the heavy playouts of Valkyria correctly
prunes playing out ladders for the loser. But sometimes in the
playouts the ladder is broken and after that there is a chance that
the stones escape a
On Mon, Sep 22, 2008 at 1:21 PM, Łukasz Lew <[EMAIL PROTECTED]> wrote:
> Hi,
>
> On Mon, Sep 22, 2008 at 17:58, Jason House <[EMAIL PROTECTED]>
> wrote:
> > On Sep 22, 2008, at 7:59 AM, Magnus Persson <[EMAIL PROTECTED]>
> wrote:
> >
> > The resu
On Sep 24, 2008, at 2:40 PM, Jacques Basaldúa <[EMAIL PROTECTED]>
wrote:
> "The approach of this paper is to treat all win rate estimations
as independent estimators with
additive white Gaussian noise. "
Have you tried if that works? (As Łukasz Lew wrote "experimental set
up would be usefu
Original Message-
From: [EMAIL PROTECTED] [mailto:computer-go-
[EMAIL PROTECTED] On Behalf Of Jason House
Sent: Wednesday, September 24, 2008 4:34 AM
To: computer-go
Subject: Re: [computer-go] MoGo v.s. Kim rematch
On Tue, 2008-09-23 at 18:08 -0300, Douglas Drumond wrote:
Attached is a quick writ
Sent from my iPhone
On Sep 24, 2008, at 5:16 PM, Jason House <[EMAIL PROTECTED]>
wrote:
On Sep 24, 2008, at 2:40 PM, Jacques Basaldúa <[EMAIL PROTECTED]> wr
ote:
Therefore, the variance of the normal that best approximates the
distribution of both RAVE and
wins/(wins
Sent from my iPhone
On Sep 27, 2008, at 10:14 AM, "Álvaro Begué" <[EMAIL PROTECTED]>
wrote:
On Fri, Sep 26, 2008 at 9:29 AM, Jason House
<[EMAIL PROTECTED]> wrote:
Sent from my iPhone
On Sep 24, 2008, at 5:16 PM, Jason House
<[EMAIL PROTECTED]>
wrote:
There's a way to create rotatable hash keys by moving nibbles around
(for 32 bit keys. Bytes for 64 bit keys)
The basic idea is to split your hash key into 8 chunks and manipulate
them similar to board rotations.
See http://housebot.wiki.sourceforge.net/0.6.3+Development
(scroll down to the
What do you mean by appeared? When were the languages inteoduced? Is
the speed and memory stuff just a claim? Or are there real benchmarks?
Sent from my iPhone
On Oct 7, 2008, at 2:38 PM, Don Dailey <[EMAIL PROTECTED]> wrote:
I have noticed there are 3 more languages that have appeared that
LLVMDC is a project to make a D compiler for LLVM. Supposedly, it is
nearly ready for prime time. I have not experimented with it yet.
My D-based bot gets 10kpps right now using gdc (D compiler for gcc).
That's fast enough to be within ~50 ELO of the best light playout bots
I've seen on CGO
Looking superficially...
The game length appears in the right ballpark. I seem to remember
110-112 moves, depending on how passed are counted.
20k playouts/core/sec seems reasonable for lightly optimized.
The center bias also looks correct.
The win rates don't look right to me. 7.5 Komi giv
You are incorrect that the following heuristics in random games lead
to finite game length:
* no eye filling
* no suicide
* no simple ko violations
Consider two eyeless chains with 3 ko's connecting them... Two taken
by black and it's white's move. Filling the one ko it has is suicide.
It m
reasonable limit such as 10.Would infinite games
still
be possible with this rule?
- Don
On Thu, 2008-10-09 at 10:25 -0400, Jason House wrote:
You are incorrect that the following heuristics in random games lead
to finite game length:
* no eye filling
* no suicide
* no simple ko violations
Co
Which multi stone capture case still exists under random games?
Sent from my iPhone
On Oct 9, 2008, at 11:12 AM, "Erik van der Werf" <[EMAIL PROTECTED]
> wrote:
On Thu, Oct 9, 2008 at 5:03 PM, Jason House <[EMAIL PROTECTED]
> wrote:
On Oct 9, 2008, at 10:41 AM, "
On Oct 9, 2008, at 11:08 AM, "Eric Boesch" <[EMAIL PROTECTED]> wrote:
On Thu, Oct 9, 2008 at 10:25 AM, Jason House
<[EMAIL PROTECTED]> wrote:
You are incorrect that the following heuristics in random games
lead to
finite game length:
* no eye filling
* no suicide
* n
, Oct 9, 2008 at 5:24 PM, Jason House <[EMAIL PROTECTED]
> wrote:
Which multi stone capture case still exists under random games?
Sent from my iPhone
On Oct 9, 2008, at 11:12 AM, "Erik van der Werf" <[EMAIL PROTECTED]
>
wrote:
On Thu, Oct 9, 2008 at 5:03 PM, Jason Hou
On Oct 11, 2008, at 2:40 PM, [EMAIL PROTECTED] wrote:
11. When selecting moves to play in the actual game (not playouts)
superko is checked and forbidden.
positional superko (as opposed to situational superko).
Whatever is used by the rules... I know CGOS and KGS Chinesee use
positional
On Oct 13, 2008, at 4:41 PM, Nick Wedd <[EMAIL PROTECTED]> wrote:
The results of yesterday's KGS bot tournament are now available at
http://www.weddslist.com/kgs/past/43/index.html
As always, I look forward to your corrections.
You give HBotSVN too much credit in the round 8 open game. Seki i
On Oct 20, 2008, at 7:28 PM, Darren Cook <[EMAIL PROTECTED]> wrote:
I will build a web site soon and hope others participate with
conforming
programs in any language or system.
I'll tentatively raise my hand to make a php version.
I'll tentatively raise my hand for a D2 version. It's a go
unit tests.
Except for speed builds, any build of my bot will run every unit test on
start up and verify a conforming implementation. game/go.d contains 20
unit tests. Along with a with other design by contact elements, I have
a total of 141 assertions.
I have the ability to draw the board stat
I have a suspect gogui will do most of what you want. Take a close
look at gtpdisplay, auto running of commands following each move, and
the various output types/display methods.
If it doesn't do what you want, it may be possible to patch it?
Sent from my iPhone
On Oct 23, 2008, at 4:38 PM,
On Oct 24, 2008, at 11:23 AM, "Erik van der Werf" <[EMAIL PROTECTED]
> wrote:
On Fri, Oct 24, 2008 at 4:57 PM, Robert Jasiek <[EMAIL PROTECTED]>
wrote:
[EMAIL PROTECTED] wrote:
In my opinion the goal of a ko rule is to prevent games from not
ending.
All restriction rules (about suicide,
On Oct 24, 2008, at 1:07 PM, Nick Wedd <[EMAIL PROTECTED]> wrote:
In message <[EMAIL PROTECTED]>, Jason
House <[EMAIL PROTECTED]> writes
On Oct 24, 2008, at 11:23 AM, "Erik van der Werf" <[EMAIL PROTECTED]
wrote:
On Fri, Oct 24, 2008 at 4:57 PM, Robert Jas
On Sat, 2008-10-18 at 09:26 -0400, Don Dailey wrote:
> I have two versions of the reference bot. A C and a Java version.
I've ported the Java position.d to D2 and added a basic wrapper around
it. Here are performance numbers on my laptop
javabot runs in 132 seconds for 1,000,000 simulations
dre
f I can get D installed and run the big
self-test too. Can you send me a statically compiled linux binary?
Absolutely... I did not want to email a half 500k attachment.
- Don
On Fri, 2008-10-24 at 23:01 -0400, Jason House wrote:
On Sat, 2008-10-18 at 09:26 -0400, Don Dailey wrote:
On Oct 25, 2008, at 1:01 AM, Don Dailey <[EMAIL PROTECTED]> wrote:
It rarely plays anything other that E5, was E6 a fluke?The other
numbers look correct to me.
I haven't posted the Vala version yet, but I bet it would be even
easier
to port from it, since Vala is heavily based on C#. I
self-test too. Can you send me a statically compiled linux binary?
>
> - Don
>
>
> On Fri, 2008-10-24 at 23:01 -0400, Jason House wrote:
> > On Sat, 2008-10-18 at 09:26 -0400, Don Dailey wrote:
> > > I have two versions of the reference bot. A C and a Java version.
>
On Oct 26, 2008, at 7:19 PM, Mark Boon <[EMAIL PROTECTED]> wrote:
One more observation, something I found curious, is that according
to the statistics twogtp put together, the average game-length
played was 119 moves. I also noticed this was the number after the
other two runs I had of 1,00
On Tue, 2008-10-28 at 08:55 +0900, Darren Cook wrote:
> Where "harder" means the gap between top programs and top human players
> is bigger, are there any games harder than go? Including games of
> imperfect information, multi-player games, single-player puzzle games.
>
> Naturally I'm most intere
You should also test your 7/8 keep heuristic to see if it's reducing
weight or avoiding the final moves that matters.
Sent from my iPhone
On Oct 28, 2008, at 3:03 PM, Don Dailey <[EMAIL PROTECTED]> wrote:
It appears from testing 3 of us are doing, that the Mark Williams
enhancement is good,
The error bars of all bots overlap. I'm not familiar enough with
BayesELO to compute p-values. I'd bet that only the 0.1 version has a
statistically significant strength difference.
Sent from my iPhone
On Oct 30, 2008, at 7:00 PM, Don Dailey <[EMAIL PROTECTED]> wrote:
The basic idea seems
I tried to connect to CGOS last night with only marginal success. It
seems that I only get one game in each time I connect.
After the first failure, I found that the client was repeatedly
reporting server connection errors. I think it was something to the
effect of "no route to host". I ass
On Sat, 2008-11-01 at 21:48 -0400, Don Dailey wrote:
> I also tried large bonuses and so far 0.20 works better than 0.10 by
> about 20 ELO after 2,500 games.I tried 0.40 and it was significantly
> weaker, so this can definitely be overdone. The final result is 131
> ELO for the 0.20 value.
, 2008-11-01 at 22:25 -0400, Jason House wrote:
I'm curious, how do you do your automated testing?
I tried using gogui-twogtp -auto with jrefbot and drefbot. When I
look
at the resulting .dat files, there's no indication of which side won
each game. I know final_status_list is i
HouseBot is ~1600 ELO on CGOS 9x9. There are also Linux binaries
available for download.
Sent from my iPhone
On Nov 4, 2008, at 10:33 AM, Eric Marchand <[EMAIL PROTECTED]> wrote:
Hi all,
Here is a list of 24 free go engines (18 with source):
http://ricoh51.free.fr/go/engineeng.htm
Please let
That's certainly true for drefbot. I think I posted about my ko issue
along with a 100 ELO strength gap. I have yet to track either down.
The twogtp discussion came from me trying to test refbots against each
other.
Sent from my iPhone
On Nov 5, 2008, at 1:13 PM, Don Dailey <[EMAIL PROTECT
I think simplistic handling of Japanese rules should play dame points
that connect chains. This avoids some problems that can arise where
ownership probability drops after the opponent plays the dame, and a
point of territory must get filled.
Even if not technically required, I can imagine
On Nov 6, 2008, at 11:09 AM, Don Dailey <[EMAIL PROTECTED]> wrote:
On Thu, 2008-11-06 at 10:44 -0500, Jason House wrote:
I think simplistic handling of Japanese rules should play dame
points
that connect chains. This avoids some problems that can arise where
ownership probability drops
On Nov 16, 2008, at 11:18 AM, "David Fotland" <[EMAIL PROTECTED]
games.com> wrote:
I thought Valkyria does local search (ladders) during the playouts.
Many Faces is lighter on the playouts. I have 17 local 3x3
patterns, then
go to uniform random without filling eyes.
No capture bias in
On Nov 18, 2008, at 7:43 AM, Michael Williams <[EMAIL PROTECTED]
> wrote:
The "well" at the end of the title is implied. And computers still
can't play 19x19 Go anywhere near the master level.
I'm not very familiar with go terms, but I think kyu means student and
dan means master.
It m
Nearly all of my early games of go were against igowin. It's a great
program that I recommend to beginners. I even got my wife into playing
it. We've both looked into buying Many Faces. Igowin is an effective
marketing strategy, even if I'm too cheap :)
I've always wondered if we'll see igo
On Nov 28, 2008, at 6:03 PM, David Silver <[EMAIL PROTECTED]> wrote:
This document is confusing, but here is my interpretation of it. And
it works well for Valkyria. I would really want to see a pseudocode
version of it. I might post the code I use for Valkyria, but it is
probably not the same
On Nov 30, 2008, at 11:49 AM, Mark Boon <[EMAIL PROTECTED]>
wrote:
Indeed, the scaling question is very important. Even though I think
I have AMAF/ RAVE working now, it's still not so clear-cut what it's
worth. With just 2,000 playouts I'm seeing a 88% win-rate against
plain old UCT tree-
On Dec 1, 2008, at 12:23 AM, Mark Boon <[EMAIL PROTECTED]> wrote:
On 30-nov-08, at 16:51, Jason House wrote:
You've claimed to be non-statistical, so I'm hoping the following
is useful... You can compute the likelihood that you made an
improvement as:
erf(# of standard d
On Dec 1, 2008, at 3:38 AM, Denis fidaali <[EMAIL PROTECTED]>
wrote:
I think it's now well known that Mogo doesn't use UCT.
I realize that i have no idea at all what Mogo do use for
it's MCTS.
There are only two things i dislike about UCT :
- It's slow to compute.
- It's deterministic
I r
I've experimented with simple stuff like pruning symmetrical moves for
the first two moves, and disallowing 1st and 2nd line moves for the
first N moves.
I toyed with the idea of rotatable zorbist hashes, but never
implemented it.
You should look up Remi's technique of progressive widen
Not really... Thomas's rules include all the typical tenuki points.
Sent from my iPhone
On Dec 11, 2008, at 9:29 AM, "steve uurtamo" <[EMAIL PROTECTED]> wrote:
the thing about "within manhattan distance (small) of other stones"
type
heuristics is that they seem to leave out the possibility o
When thinking about the apparent strength loss, I came up with a
potential theory: consistency. With more simulations, noise has less
of an impact. I'm going to guess that the known bias of AMAF leads to
blunder that is played more consistently. Bots with fewer simulations
would make the bl
I hope you're joking...
Sent from my iPhone
On Dec 30, 2008, at 2:01 PM, Don Dailey wrote:
On Tue, 2008-12-30 at 12:52 -0500, steve uurtamo wrote:
that's with "or manhattan distance 2" as well? how about 3 or 4?
It looks like 3 is no good:
Rank Name Elo+- games score oppo. draw
Look at the board position at move 77 (black C9). The move of Black J3 is
both a violation of positional superko (used by CGOS) and the more lax
situational super ko.
On Fri, Jan 9, 2009 at 12:40 PM, Mark Boon wrote:
> The attached game played on CGOS was awarded a win for white due to an
> ill
On Jan 26, 2009, at 6:26 PM, matt harman
wrote:
> That the missunderstanding right there.
> 1 child will be chosen and 1 simlation will be run.
Thanks for the quick answer, so 1 simulation is run because too many
will give lots of noise to the result?
Just the opposite. The noise in the w
How many playouts per second do you get with each version?
Sent from my iPhone
On Feb 1, 2009, at 4:46 PM, "Isaac Deutsch" wrote:
By the way, I got about 75 ELO points (1650->1720) with light
playouts out of RAVE. Do you think this is in the expected range?
It's not really similar to the 2
On Feb 2, 2009, at 6:57 AM, "Isaac Deutsch" wrote:
Hi Issac,
You should be more in the range of +200-300 ELO, at least with
pattern
based
playouts.
Sylvain
Isaac. They are not pattern based playouts, but as I said uniformly
random.
I reckon the effect of RAVE is less with these?
"Ho
On Feb 2, 2009, at 9:40 AM, Jason House
wrote:
On Feb 2, 2009, at 6:57 AM, "Isaac Deutsch" wrote:
Hi Issac,
You should be more in the range of +200-300 ELO, at least with
pattern
based
playouts.
Sylvain
Isaac. They are not pattern based playouts, but as I said uniforml
On Feb 2, 2009, at 12:09 PM, "Isaac Deutsch" wrote:
Wow, thanks for all the answers! You're being really helpful.
"Do you use UCT with a too large exploration term?"
That's a good idea. I actually use a rather big value for c=0.5. I
might try
lowering it. Thanks! (Precisely, the formula is
On Sun, Feb 1, 2009 at 4:46 PM, Isaac Deutsch wrote:
> By the way, I got about 75 ELO points (1650->1720) with light playouts out
> of RAVE. Do you think this is in the expected range? It's not really similar
> to the 20%->60% win rate rise vs. GnuGo described in some papers...
My bot is about
On Tue, Feb 3, 2009 at 1:53 PM, Isaac Deutsch wrote:
> Hi Jason,
>
> Thanks for your numbers. I might try to limit my bot to 50k playouts and 1
> core, but I usually simulate as long as time permits.
That kind of setup should make it easier to compare. There have been a few
times in the past w
On Feb 3, 2009, at 7:08 PM, Darren Cook wrote:
The server could also run traceroute before and during the game to
get a
fair idea of what is reasonable net lag for that particular client.
Couldn't traceroute also be used with server-side timekeeping? The
server could credit the player for
On Feb 6, 2009, at 12:55 PM, "Isaac Deutsch" wrote:
The rating of the bot still seems to be drifting upwards, but I
think I can
conclude my UCT implementation is OK afterall. Many thanks to the bots
provided. Does someone have a bot that does 50k light playouts +
RAVE? I
would be most grat
On Fri, 2009-02-06 at 18:55 +0100, Isaac Deutsch wrote:
> The rating of the bot still seems to be drifting upwards, but I think I can
> conclude my UCT implementation is OK afterall. Many thanks to the bots
> provided. Does someone have a bot that does 50k light playouts + RAVE? I
> would be most g
On Feb 10, 2009, at 12:42 PM, Olivier Teytaud
wrote:
Time must have been with about 40 seconds per move
in the average.
MoGo plays in 30s on the client machine - the 40s for KGS
is for ensuring that there's no problem with net lag.
Info on the game in chinese:
http://www.cdnews.com.tw/cdn
On Feb 16, 2009, at 12:03 AM, Markus Enzenberger > wrote:
Unfortunately, I am leaving the team in a few weeks, and it is
unclear how much I can contribute after that. Martin is usually too
busy for doing maintainer work and at the moment there is only
funding for a few months for a new pr
On Mon, Feb 16, 2009 at 9:57 AM, Jason House wrote:
>
> I took a serious look at Fuego a few months back. The code appeared to use
> modern C++ libraries, but also showed its age/lineage. If I remember right,
> the Fuego source comes with 3 projects that all depend on each other. I
&g
I'd be more than happy to work with you and the other members of your
group. I'm getting close to wrapping up a restructuring of my bot that
allows easily swapping out evaluation methods and search techniques.
As an example, here's the code that does a few basic MC searches:
11 static if (s
On Feb 17, 2009, at 12:55 PM, Dave Dyer wrote:
While your goal is laudable, I'm afraid there is no such thing
as a "simple" tree search with a plug-in evaluator for Go. The
problem is that the move generator has to be very disciplined,
and the evaluator typically requires elaborate and expens
On Feb 17, 2009, at 4:39 PM, wrote:
I've been looking into CGT lately and I stumbled on some articles
about approximating strategies for determining the sum of subgames
(Thermostrat, MixedStrat, HotStrat etc.)
Link?
It is not clear to me why approximating strategies are needed. What
i
I'm glad to hear they're back. They provide a good opportunity to chat
with other developers as well as a bit of extra motivation to do
development.
Sent from my iPhone
On Mar 28, 2009, at 8:16 AM, Nick Wedd wrote:
After a break, I am resuming KGS bot tournaments. I have not yet
decided
Which Mogo paper(s)? Not all Mogo papers contain ideaa used by Mogo in
public releases / competitions. Some things are just research. I
remember hearing that the grandfather heuristic isn't good.
Sent from my iPhone
On Mar 30, 2009, at 5:36 PM, Matthew Woodcraft
wrote:
How are transpos
I think you're looking for a post by Łukasz Lew about the epsilon
trick...
Sent from my iPhone
On Mar 31, 2009, at 8:37 PM, Michael Williams > wrote:
It seems like there was a short discussion here recently about a
strategy for reducing the amount of time spent updating the MCTS
search t
I disagree with your comment about AyaMC's move 28 in round 4. The
move looks to me like it primarilly aims to build a large framework
along the bottom/center. All basic fights seem to favor white..
I'm only KGS 3k, so take my comments with a grain of salt...
Sent from my iPhone
On Apr 6, 2
This reminds me of a related question I had a while back. In a single
MCTS/UCT search node, how do people store the children? Does a node
contain summaries of all their children, or just pointers to the
children?
Pure pointers are simple but requires allocating many more objects,
includin
On Apr 8, 2009, at 3:15 AM, Łukasz Lew wrote:
On Tue, Apr 7, 2009 at 23:52, Claus Reinke
wrote:
Last time I looked more closely at what my MC bot (simple, no tree)
was doing, I noticed that it has a tendency to try the impossible
moves
(suicides) first, discarding them as impossible for e
That sounds like a classic _positional_ super ko violation. Any board
repetition is a ko violation, regardless of the player to play.
Sent from my iPhone
On Apr 13, 2009, at 9:32 AM, "Brian Sheppard"
wrote:
Black is flagged for an illegal Ko at the end of game 738921 on
CGOS. Black play
I use atomic increments and atomic reads. It's really simple x86
assembly. To do that, I used to have a counter for wins and a total
simulations counter, but switched to wins and losses counter. Doing
that allows independent increments to those counters.
I have not done a lockless hashtable
Please take the ensuing rules argument/discussion off-list. The last
ko rules discussion resulted in way too many e-mails in everyone's
inbox.
Sent from my iPhone
On Apr 14, 2009, at 2:06 PM, Robert Jasiek wrote:
Richard Brown wrote:
And what is the _reason_ to leave out the information o
On Apr 14, 2009, at 5:42 AM, Rémi Coulom
wrote:
Jason House wrote:
I use atomic increments and atomic reads. It's really simple x86
assembly. To do that, I used to have a counter for wins and a total
simulations counter, but switched to wins and losses counter. Doing
that a
On Apr 14, 2009, at 6:46 PM, Rémi Coulom
wrote:
Jason House wrote:
In my implementation, I found that node allocation is the most
difficult part. For a tree, I suppose it may be done easily by pre-
allocating a node pool for each thread, and managing memory
allocation locally.
I
On Apr 14, 2009, at 7:57 PM, Rémi Coulom
wrote:
Jason House wrote:
Out of curiosity, how do you intelligently delete old nodes?
Reference counting won't always work due to cycles, and a nieve
scan of the tree could block all threads.
I store a date of birth in every node. A
ss about what reusing a hash entry could mean besides replacing
a complete slot in the table.
Jason House wrote:
On Apr 14, 2009, at 7:57 PM, Rémi Coulom fr> wrote:
Jason House wrote:
Out of curiosity, how do you intelligently delete old nodes?
Reference counting won't always wo
I can run my bot (1600-1700 ELO), but I may a few days depending on
free time over the weekend.
Sent from my iPhone
On Apr 17, 2009, at 4:23 PM, "Brian Sheppard"
wrote:
There are actually very few programs playing on CGOS (9x9). My
engine is rated around 1000, which means that it plays
From memory, when something goes wrong (engine crash?), the client
incorrectly sends the previous move. This then causes the server to
boot the bot due to an illegal move. Everything then shuts down,
leaving a confused programmer :)
Sent from my iPhone
On Apr 18, 2009, at 3:51 PM, "Brian S
00 for 50k.
I messed up the first round and both bots lost on time due to no
internet connection.
Sent from my iPhone
On Apr 17, 2009, at 4:49 PM, Jason House
wrote:
I can run my bot (1600-1700 ELO), but I may a few days depending on
free time over the weekend.
Sent from my iPhone
.
I will try to run it today again.
Lukasz
2009/4/20 Jason House :
I've started two bots: hb797-10k and hb797-50k
They are pure UCT+RAVE with light playouts, one search thread, and no
pondering.
The number at the end represents the playouts per move.
I forget the approximate ratings for
101 - 200 of 660 matches
Mail list logo
