Re: [Computer-go] CGOS boardspace

[Christoph Birk]

On 11/19/2015 05:46 AM, Joshua Shriver wrote:

I did a restart of the 9x9 and 19x19 as a test. Anyone mind testing it
to see if you can connect?


I connected two bots to
  cgos.boardspace.net:6867
but no games are starting and the page
  http://cgos.boardspace.net/9x9/standings.html
does not update either.

Christoph


___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] Komi 6.5/7.5

[Christoph Birk]

On Nov 5, 2015, at 4:44 AM, Nick Wedd  wrote:
> However, there's a powerful counterargument to the above  I can put the first 
> black stone on the board as well as any professional can. And now, assuming I 
> am playing an equally weak human, it's White who suffers most from the 
> imperfection of our subsequent moves.

But White already got the komi ….
Christoph


___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] CGOS

[Christoph Birk]

On Jun 10, 2015, at 9:53 AM, Detlef Schmicker d...@physik.de wrote:
 After my ISP crashed, I do not get up 9x9 at the moment.
 Immediatly myCtest tries to connect from within the middle of a game i
 think and DODs the server….

They try to re-connet once a minute … I stopped them now.
Christoph



Once is happenstance. Twice is coincidence. The third time it's enemy action. 
-- Auric Goldfinger

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] CGOS

[Christoph Birk]

On 05/26/2015 02:41 AM, Detlef Schmicker wrote:

-BEGIN PGP SIGNED MESSAGE-
it should be up nearly 24/7 I hope and use less than 5W electrical
power, until the sd card is full :)


Thank you,
Christoph


___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] CGOS

[Christoph Birk]

On May 23, 2015, at 12:40 AM, Detlef Schmicker d...@physik.de wrote:
 24/7 is only useful, if other than open source bots are run on the
 server, otherwise the author can run it simply on gomill...

While I agree that it is not ideal having so few programs running,
shutting down the server is even worse, or not?

Christoph

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] CGOS

[Christoph Birk]

On May 22, 2015, at 10:46 AM, Detlef Schmicker d...@physik.de wrote:
 I wonder, if it would help to put it up once a week or so, with announcement, 
 and take it down again, if the number of bots falls below 5 or so?

I am not actively developing a bot, but IMHO without being up
24/7 CGOS is not very useful.

Christoph

---
Science advances one funeral at a time -- Max Planck



___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] CGOS

[Christoph Birk]

On May 1, 2015, at 10:21 PM, Detlef Schmicker d...@physik.de wrote:
 I set up a CGOS server at home. It is connected via dyndns, which is not 
 optimal of cause :(

Great, I will try to run ‘myctest’ on Monday,
Christoph

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] What's a good playout speed?

[Christoph Birk]

On Apr 7, 2015, at 4:34 AM, Urban Hafner cont...@urbanhafner.com wrote:
 I suspected you'd say something like this. ;) It is definitely on my list of 
 things to steal a few things from Michi. But maybe I'll start with simpler 
 and/or well defined things like RAVE or the hand picked MoGo 3x3 patterns. 
 That way it's easy to see if I really screwed something up. The bot is still 
 rather weak so adding some of those features should really improve the 
 strength.

How many playouts (per move) does 'stop_0.9-2b’ do?

Christoph



___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] What's a good playout speed?

[Christoph Birk]

On Apr 7, 2015, at 7:16 AM, Urban Hafner cont...@urbanhafner.com wrote:
 I wouldn't know, Christoph. My (and Igor's) bot is called Iomrascálaí. :P 
 It's running as the various Imrscl-XYZ bots on CGOS due to the username 
 length restriction and the fact that the current CGOS can't handle Unicode 
 characters.
 This bot however does around 4.5k pips on 9x9 and 1k apps on 19x19 running on 
 a 2,2 GHz Intel Core i7 (6 months only MacBook pro). The versions on CGOS run 
 using 8 threads and I get a speedup of about 4.5x.

thanks, I agree 1400 is about as far as simple UCT will get you.
My simple UCT implementation (myCtest-xxk-UCT) gets about
1200, but it does not do any adjustments to the number of playouts
per move depending on the time remaining, so I have to limit
it to 40k playouts per move.
Have you thought about using the partial tree of the previous move
as a bias?

Chrisoph


___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] Weak bots to run on CGOS

[Christoph Birk]

On Mar 20, 2015, at 5:11 AM, Urban Hafner cont...@urbanhafner.com wrote:
 So, I now have a new version of my bot running on CGOS 
 (http://cgos.boardspace.net/13x13/cross/Imrscl-016-AMAF.html). It's still 
 considerably weaker than GnuGo so I'm pretty sure it will loose all games 
 against it. However, it's now much stronger than any other bot running on 
 CGOS and I guess it will be hard to get a good rating. Are there any bots 
 that I could run that are weaker than GnuGo, but not that much weaker?

I started myCtest-20k-UCT (BayesELO=1187), and myCtest-40k-UCT (1343).
If necssary I will run an -80k version instead of the -20k.

Christoph

ps. If someone stopped ‘Stop-08’, ’Stop-05’ and ‘resign13’ we would get 
more useful games.

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] Weak bots to run on CGOS

[Christoph Birk]

On Mar 9, 2015, at 2:08 AM, Urban Hafner cont...@urbanhafner.com wrote:
 I'm currently running Brown (random bot) and GnuGo on CGOS 13x13. Mainly to 
 get a feel for the strength of my own bot. And my bot is really bad. ;) So 
 bad that it looses all games against GnuGo, but wins all games against Brown. 
 So, the rating is a bit useless I assume as there are no bots that are in 
 strength between GnuGo and the random player. Are there any bots in that 
 range out there? I'd be willing to run them myself on CGOS.

myCtest-xxk is a pure random player.
myCtest-xxk-UCT adds a tree, nothing else.

Christoph

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] Weak bots to run on CGOS

[Christoph Birk]

On Mar 9, 2015, at 7:50 AM, Christoph Birk b...@obs.carnegiescience.edu wrote:
 On Mar 9, 2015, at 2:08 AM, Urban Hafner cont...@urbanhafner.com wrote:
 I'm currently running Brown (random bot) and GnuGo on CGOS 13x13. Mainly to 
 get a feel for the strength of my own bot. And my bot is really bad. ;) So 
 bad that it looses all games against GnuGo, but wins all games against 
 Brown. So, the rating is a bit useless I assume as there are no bots that 
 are in strength between GnuGo and the random player. Are there any bots in 
 that range out there? I'd be willing to run them myself on CGOS.
 
 myCtest-xxk is a pure random player.

I meant to write pure MC player.

 myCtest-xxk-UCT adds a tree, nothing else.

Christoph

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] Weak bots to run on CGOS

[Christoph Birk]

I would like to ask the owner of 'resign13' to stop it, please.
Since the rating algorithm appears to be capped at '0' Elo,
'resign13' is skewing the ratings at the lower end.

Thanks,
Christoph

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] cgos.computergo.org down?

[Christoph Birk]

On Mar 2, 2015, at 6:55 AM, Joshua Shriver jshri...@gmail.com wrote:
 It was migrated back to the original boardspace.  Please try there.
 cgos.boardspace.net

The 9x9 and 19x19 servers never got running. The 13x13 server
ran for a while, but crashed about a month ago and has not been
restarted.

Christoph

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] CGOS back online

[Christoph Birk]

On Jan 16, 2015, at 1:51 AM, valky...@phmp.se wrote:
 I forgot to turn of automatic Power off in Windows so after an hour my 
 computer hibernated. I had started Valkyria again this morning (now using 6 
 threads) and then CGOS seemed to recover.
 
 Maybe CGOS froze because of this?

No, CGOS kept running fine after Valkyra disconnected,
Christoph

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] CGOS back online

[Christoph Birk]

On 01/16/2015 12:03 PM, David Doshay wrote:

cgos.boardspace.net http://cgos.boardspace.net says:
At the current time there is one player called FatMan with a fixed ELO
of 1800 on the 9x9 server and Gnugo-3.7.10 at level 10 serves as the
anchor player on the 13x13 and 19x19 server, also with a fixed ELO of 1800.


Should we use Gnugo-3.7.10 as the anchor for 9x9 too?
It was rated 1858.6 by 'bayeselo'.

Christoph

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] CGOS back online

[Christoph Birk]

On Jan 15, 2015, at 9:35 AM, Joshua Shriver jshri...@gmail.com wrote:
 Aye I'm still tinkering with it, and trying to get anchors on.  Still
 having issues. :(

The 13x13 server is up and running,
Christoph

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] CGOS back online

[Christoph Birk]

On Jan 15, 2015, at 1:03 AM, Urban Hafner cont...@urbanhafner.com wrote:
 On Wed, Jan 14, 2015 at 5:45 PM, Christoph Birk 
 b...@obs.carnegiescience.edu wrote:
 http://cgos.boardspace.net/9x9/standings.html
 was updated last about 2 years ago.
 
 I noticed that, too. Also, it seems like there are still games in progress 
 from 2012 which is rather unlikely. So it seems like it's not quite up and 
 running, yet. 

The 13x13 and 19x19 ‘standings’ pages have a recent date, but are empty.
Christoph

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] CGOS back online

[Christoph Birk]

On Jan 14, 2015, at 7:30 AM, folkert folk...@vanheusden.com wrote:
 I've connected a couple of programs but nothing happens.
 They login and that's about it.

Same here.

http://cgos.boardspace.net/9x9/standings.html
was updated last about 2 years ago.

Christoph

___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] alternative for cgos

[Christoph Birk]

On Jan 13, 2015, at 4:52 AM, Woody Folsom woody.fol...@gmail.com wrote:
 I would be interested in participating, particularly as a containerized 
 environment puts me on a more even footing with projects which have a lot 
 more hardware to throw at the problem.

That’s an interesting setting for a tournament but not really a replacement
for CGOS, which was mostly used for testing.

Christoph





___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] Teaching Deep Convolutional Neural Networks to Play Go

[Christoph Birk]

On 12/15/2014 01:39 PM, Dave Dyer wrote:


You don't need a neural net to predict pro moves at this level.

My measurement metric was slightly different, I counted how far down the
list of moves the pro move appeared, so matching the pro move scored
as 100% and being tenth on a list of 100 moves scored 90%.


There is a huge difference between matching a pro move or have
it #10 on a list of 100


Combining simple metrics such as 3x3 neighborhood, position on the board,
and proximity to previous play, you can easily get to an average score
of 85%, without producing noticeably good play, at least without a search
to back it up.


85% is basically meaningless, I am sure even a mid-kyu player
can put a pro-move in the top 15% of 100 moves.

Christoph




___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Re: [computer-go] Dynamic komi at high handicaps

[Christoph Birk]

On Aug 12, 2009, at 2:51 PM, Don Dailey wrote:
I disagree.   I think strong players have a sense of what kind of  
mistakes to expect, and try to provoke those mistakes.   Dynamic  
komi does not model that.


It also does the opposite of making the program play provocatively,  
which I believe is necessary to beat a weaker player with a large  
handicap against you.Instead of making it fight,  it encourages  
the program to be content with less.   How does this model strong  
handicap players?


Maybe dynamic komi works better for BLACK? Computers are still
a looong way from actually _giving_ a handicap.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Dynamic komi at high handicaps

[Christoph Birk]

On Aug 12, 2009, at 3:10 PM, Don Dailey wrote:
If the handicap is fair,  their chance is about 50/50.   However,   
rigging komi to give the same chance is NOT what humans do.   The  
only thing you said that I consider correct is that humans estimate  
their chances to be about 50/50.
One thing humans do is to set short term goals and I think dynamic  
komi is an attempt to do that - but it's a misguided attempt  
because you are setting the WRONG short term goal.


Setting the komi to that the game is 50/50 creates the (correct)
short term goal of gaining a few points, then again, and again ...

Christoph
 
___

computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Dynamic komi at high handicaps

[Christoph Birk]

On Aug 12, 2009, at 3:43 PM, Don Dailey wrote:
I believe the only thing wrong with the current MCTS strategy is  
that you cannot get a statistical meaningful number of samples when  
almost all games are won or lost.You can get more meanful  
NUMBER of samples by adjusting komi,  but unfortunately you are  
sampling the wrong thing - an approximation of the actual goal.
Since the approximation may be wrong or right,  your algorithm is  
not scalable.   You could run on a billion processors sampling  
billions of nodes per seconds and with no flaw to the search or the  
playouts still play a move that gives you no chances of winning.


I think you got it the wrong way round.
Without dynamic komi (in high ha
ndicap games) even trillions of simulations
with _not_ find a move that creates a winning line, because the is none,
if the opponet has the same strength as you.
WHITE has to assume that BLACK will make mistakes, otherwise there
would be no handicap.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Dynamic komi at high handicaps

[Christoph Birk]

On Aug 12, 2009, at 10:31 PM, Petri Pitkanen wrote:

Maybe they are long way from giving handicaps to you. But best of bots
in KGS are around 2k and there are hundreds of  9k and weaker players
present there at all times. So being able to play white is worthy
thing at least for commercial bot.


That's correct. I have a more academic point of view.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Finding specific CGOS game

[Christoph Birk]

On Aug 2, 2009, at 8:05 AM, Don Dailey wrote:
Here are last few games of Pebbles where pebbles lost on time as  
black - which is what would happen in a crash.


Pebbles is losing a lot of games on time.


And all of them as black.

794069|gnugo-3.7.12-l10F|1759|Pebbles|2155|2009-06-23 12:51|23130| 
306264|W+Time|y
796644|fuego-0.4-slow|2050|Pebbles|2144|2009-06-28 07:21|213889| 
314234|W+Time|y
797185|Nomitan_010|2099|Pebbles|2145|2009-06-29 05:25|166424|457572| 
W+Time|y
798339|fuego-0.4-slow|2062|PebblesToo|2120|2009-07-01 10:00|285864| 
300965|W+Time|y
800546|Aya681_1sim|2260|PebblesToo|2144|2009-07-05 07:03|42305| 
307156|W+Time|y
803518|Aya681_1sim|2245|Pebbles|2209|2009-07-09 22:44|8332| 
302132|W+Time|y
804223|Nomitan_tanabata|2102|Pebbles|2221|2009-07-11 06:16|286093| 
365965|W+Time|y
805600|PebblesToo|2210|Pebbles|2234|2009-07-13 06:08|156681|313945|W 
+Time|y
809161|fuego-0.4-slow|2015|PebblesToo|2257|2009-07-17 09:13|269777| 
300240|W+Time|y
811229|lingo-B5.10|2116?|Pebbles|2259|2009-07-19 02:55|174175| 
303600|W+Time|y
811242|fuego-0.4-slow|2002|PebblesToo|2227|2009-07-19 03:04|0| 
312380|W+Time|y
813140|gnugo-3.7.12-mc|1940|PebblesToo|2204|2009-07-20 14:26|0| 
314432|W+Time|y
813727|UmeBot-1b|1433|Pebbles|2226|2009-07-21 00:42|116508|305905|W 
+Time|y
816181|Fuego4C4PlaPo20Mno|2395|PebblesToo|2205|2009-07-22 21:10|0| 
314367|W+Time|y
819178|Aya681_1sim|2194|Pebbles|2223|2009-07-25 01:51|17213| 
300872|W+Time|y
820646|GnuGo-mc-10K-lev11|2013|PebblesToo|2195|2009-07-26 01:57| 
56697|306864|W+Time|y

821628|GG-500|1738|Pebbles|2210|2009-07-26 17:41|6380|310153|W+Time|y
823871|TakeRaveGom_ct1_15|2074?|PebblesToo|2186|2009-07-28 04:44| 
260261|308616|W+Time|y
824478|gnugo-3.7.12-mc|1899|Pebbles|2211|2009-07-28 15:35|164822| 
303483|W+Time|y
824905|Fuego4C4PlaPo20Mno|2375|PebblesToo|2190|2009-07-28 22:40| 
148900|306979|W+Time|y
824920|TakeRaveGom_ct1_15|2081|Pebbles|2200|2009-07-28 22:50|0| 
301031|W+Time|y
825424|GnuGo-mc-10K-lev11|2020|PebblesToo|2189|2009-07-29 07:04| 
276431|301032|W+Time|y
825563|Pebbles|2194|PebblesToo|2183|2009-07-29 09:27|232670|302187|W 
+Time|y
830994|GnuGo-mc-10K-lev11|1977|PebblesToo|2191|2009-08-02 07:05| 
255392|303189|W+Time|y


Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: Mirror Go against Zen

[Christoph Birk]

On Thu, 23 Jul 2009, Zach Wegner wrote:

White can simply pass if black plays in the center. Black passing in
response would be an instant loss (provided komi is  0 of course).


Quite the opposite. If white passes after black's first move
since all empty points just touch black, so black get the
entire board as territory.

Christoph


___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] New CGOS - need your thoughts.

[Christoph Birk]

On Mon, 15 Jun 2009, Brian Sheppard wrote:

Please don't do anything that decreases the frequency of games in order
to accommodate programs that want to play on multiple venues. Keep venues
strictly separate. Programs that want to play on multiple venues can just
log in multiple times.


I second that opinion.
If there is a second venue, I'd prefer longer time controls.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Rating Drift

[Christoph Birk]

On Tue, 21 Apr 2009, sheppar...@aol.com wrote:

Pebbles learns from every game it plays. So I can't agree; drift is
inherent.


But since you had bugs in the earlier version, how do you know,
without restarting it after bug-fixes how much of the drift
is from the learning part and how much from the bug-fix?

Even for a learning program it might be a good idea to change
the name by adding a version number after bug-fixes or major
improvements.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Rating Drift

[Christoph Birk]

On Tue, 21 Apr 2009, Jason House wrote:
AMAF and RAVE are the same thing. The MoGo team pioneered use of AMAF but 
called it RAVE because of their paper's target audience.


I always thought them to be the application of the same heuristic at
a different time.
AMAF is usually applied at the end of the search, while RAVE
guides the search. But maybe that's just me :-)

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Could be that nobody is playing?

[Christoph Birk]

On Mon, 20 Apr 2009, ?ukasz Lew wrote:

Is there a rating drift? I remember that pure UCT no RAVE with 100k
playouts got over 1700 elo.


That seems a little high. My 50k-pure-UCT searcher is around
1580 for a long time.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Could be that nobody is playing?

[Christoph Birk]

On Mon, 20 Apr 2009, ?ukasz Lew wrote:

Is there a rating drift? I remember that pure UCT no RAVE with 100k
playouts got over 1700 elo.


There is no 'anchor' (FatMan-1 ?) runnig on CGOS-9x9 for at
least 36 hours. That could create a drift.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Could be that nobody is playing?

[Christoph Birk]

On Fri, 17 Apr 2009, Brian Sheppard wrote:

I saw on Sensei's Library page http://senseis.xmp.net/?CGOSBasicUCTBots that
there are a range of basic UCT implementations that would be excellent
opponents (rating 1171 through 1603), but I haven't seen these players in
weeks. Is it possible to get them back up?

If so, I would deeply appreciate it.


I have started the following standards again:

myCtest-10k: no tree, no heuristics, 10k light playouts.

myCtest-10k-UCT: simple UCT-tree searcher, 10k light playouts.

myCtest-10k-AMAF: no tree, 10k light playouts, uses the
  'all moves as  first' heuristic for move selection.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] How to properly implement RAVE?

[Christoph Birk]

On Feb 6, 2009, at 9:55 AM, Isaac Deutsch wrote:
By the way, I've seen 2 games when checking my bot's status where  
one of the
myCtest bots lost because of an illegal ko move. Maybe there's a  
bug in

handling superko?


Not a bug, I never implemented it :-(

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] UCT concept

[Christoph Birk]

On Mon, 26 Jan 2009, matt harman wrote:

With an empty board, assuming I am using proximity heuristic of 1 Manhattan 
distance,
from the root I will have 4 possible positions which will make up 4 children of 
the root.

Each child will be simulated (eg) 1000 times and a winrate is calcuated.
If child A has the highest winrate it will be exploited due to UCB1.


That the missunderstanding right there.
1 child will be chosen and 1 simlation will be run.


Because you have reached the leaf you will create 4 more children for A,
and this gets repeated.

My Question is: at each level of the tree do you return back to the root and 
traverse through the
arm again?


yes, after each simulation.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

RE: [computer-go] UCT concept

[Christoph Birk]

On Mon, 26 Jan 2009, matt harman wrote:

Thanks for the quick answer, so 1 simulation is run because too many
will give lots of noise to the result? if only 1 is run then the 4 children can 
either win or lose
the single simulation 0 or 1. This would be non-deterministic so how would you
decide which child to exploit?


For every pass through the tree (simulation) you use the UCT
formula at every node to determine which branch (child) to choose.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] 3-4-5 rule

[Christoph Birk]

On Tue, 30 Dec 2008, Don Dailey wrote:


On Tue, 2008-12-30 at 14:23 -0500, Jason House wrote:

I hope you're joking...


It lost twice as many as it won, you're not convinced?  :-)

Ok,  I'll let it run a few hundred more games just in case it somehow
manages to turn things around.


I agree with Jason ... how can it (distance=3) be worse?

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] 3-4-5 rule

[Christoph Birk]

On Tue, 30 Dec 2008, Don Dailey wrote:

Distance 3 could easily play worse - we shall see.   Just because a
distance 3 move is sometimes good doesn't mean it will make the program
play better not throwing those out.   If it's RARELY best, then the
reduced effort and increased focus on (usually) more relevant moves
could be a win.   In fact I expect distance 2 to be better for that
reason.


IMHO 'd3' could be worse than 'd2' but not worse than 'base'.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: Opportunity to promote ...

[Christoph Birk]

On Nov 18, 2008, at 11:28 AM, [EMAIL PROTECTED]  
[EMAIL PROTECTED] wrote:
It depends very much on what exactly you mean by amateur master  
level. Is it a level that compares to amateur master level in chess?
And what is amateur master level in chess? USCF master, FIDE master  
or international master?
Some time ago I participated in a discussion about comparing chess  
titles to go ranks by evaluating effort, prestige and other factors  
[1].


In my opinion:
1: USCF master compares to about 4d
2: FIDE master compares to about 6d
3: International master compares to about 7d/1p


I suggest to overlay the histrograms of player ratings and shift one
along the x-axis until the mode (peak in the distribution) are at the
same point. From that it should be easy to compare chess ELO
ratings with go ratings (or ranks).

Christoph
 
___

computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

RE: [computer-go] Re: Opportunity to promote ...

[Christoph Birk]

On Wed, 19 Nov 2008, [EMAIL PROTECTED] wrote:

I think that would not be enough, because that would only fix one point.


You can use the width too. That should give a pretty good comparision
for moderatly strong/weak players (see below).

EGF ratings are not pure Elo ratings. EGF ratings are weighted to fit 
100 points for one handicap stone, which happens to match about 65% 
winning percentage in even games for medium level players (around 3k).


That should not matter much. The typical chess player should be
as strong as the typical Go player and I also expect the strength
distribution to follow similar lines.

Also, I am not aware that there exists a histogram of the worldwide go 
population.


Why would you need world-wide data? Use US-Go/Chess or European data.
The lack of pros in this distributions should not matter much, as these
are very few at the top end of the distribution.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Another enhancement to AMAF

[Christoph Birk]

On Wed, 29 Oct 2008, Mark Boon wrote:
the implementation with one that clears the array instead of increasing the 
marker. And I'll only have to make changes in one place instead of dozens, or 
more. Not that I had this in mind when I designed it, it's just the 
beneficial side-effect of OO design in general.


[troll on]
What's that todo with OO design?
You can do the same by writing a function (eg. in 'C').
[troll off]

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] reference bots java and C

[Christoph Birk]

On Mon, 20 Oct 2008, Don Dailey wrote:

And now after about 11,000 games we are within 1 standard deviation and
the score is very close to 50% so I have confidence that we have 2
functionally equivalent bots.


Why are they not running on CGOS?

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] reference bots java and C

[Christoph Birk]

On Mon, 20 Oct 2008, Don Dailey wrote:


On Mon, 2008-10-20 at 13:47 -0700, Christoph Birk wrote:

On Mon, 20 Oct 2008, Don Dailey wrote:

And now after about 11,000 games we are within 1 standard deviation and
the score is very close to 50% so I have confidence that we have 2
functionally equivalent bots.


Why are they not running on CGOS?


I can run 11,000 games in a day or so with my own tester,  it would take
forever to run that many games on CGOS.


Sure, that's why I asked AFTER you were confident that they
were equal :-)


Nevertheless, for the past hour or two I have both running on CGOS and
plan to let them get a lot of games in.


Great, let's see what the standard ELO rating turns out to be.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] komi for 9x9

[Christoph Birk]

On Thu, 9 Oct 2008, Ingo Althöfer wrote:

I would like to see all Go programs to be able to live with
possible draws (or even with any score spectrum).


My program (myCtest) works with draws, but it's fairly weak
at about 1550 ELO (3.2 GHz P4).

Christoph


___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] AMAF Scalability study + Responses to previous

[Christoph Birk]

On Thu, 9 Oct 2008, Denis fidaali wrote:
tCan we degrade performances more with more simulations ? :) How 
does 5000AMAF fares
agains 1AMAF, i wonder. Although i'm more interested about the 
upscales that the downscales :)


I tried 50k vs 10k and saw no further improvement (no degradation
either).

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Light simulation : Characteristic values

[Christoph Birk]

On Wed, 8 Oct 2008, Don Dailey wrote:

Christoph,
Do you use all-moves-as-first?   If not, this data seems to match mine
very well.   The upper bound seems to be around 1300 ELO give or take a
few ELO.Ike seems to be around 1300 ELO with 10k play-outs but they
are all-as-first.I'll let it run a few days.


'myCtest-xxk' just uses light playouts and no further improvement.

'myCtest-10k-AMAF' uses 'all-moves-as-first'

'myCtest-xxk-UCT builds UCT tree.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] 7.5-komi for 9x9 in Beijing

[Christoph Birk]

On Wed, 8 Oct 2008, Don Dailey wrote:

much more common.There were just a few games that used 6.5 komi
because when I first started CGOS I had set 6.5 by mistake but I think
that was just for a few hours at most.   The vast majority of these are
7.5 komi games:


After all this discussion about komi for 9x9 games, wouldn't you
think that using 7.5 was a mistake and go back to 6.5 ?

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] More Characteristic values (AMAF Characteristics from empty board)

[Christoph Birk]

On Wed, 8 Oct 2008, Denis fidaali wrote:
Now, i wanted to make sure that my implementation had any chances to be 
correct. So i though I'd post the characteristic statistical values that 
i get out of it. Indeed i though it could benefits others later on, in 
particular if someone could corroborate them :)


As I said earlier: let it run on CGOS and play 200 games ...
You should get around 1000 ELO using 10k simulations, or
1300 ELO with 50k.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] More Characteristic values (AMAF Characteristics from empty board)

[Christoph Birk]

On Wed, 8 Oct 2008, Denis fidaali wrote:

To Don and Christoph : I reallize that i was probably not as clear as i though 
i was.
I have built up a light simulator. There are no tree involved. It is 
only choosing a move with equiprobabilty from the set of empty points on 
the board.


That's exactly what 'myCtest-xxk' is doing.

Only the 'myCtest-xxk-UCT' builds a tree.

Christoph


If the move is not valid, it just choose another one. If it's a pseudo 
eye then it again chooses another point. Whenever no point can be 

validly chosen, it just pass. Whenever two consecutives pass occurs, the 
simulation end.


Now, i wanted to make sure that my implementation had any chances to be 
correct. So i though I'd post the characteristic statistical values that i get 
out of it. Indeed i though it could benefits others later on, in particular if 
someone could corroborate them :)


So here is another set of Values. It is the all-move-as-first score from an 
empty board for black stones (black plays first in my simulations). It gives 
the probability of black wining the game (while playing randomly), if he has 
played an intersection before white during the simulation.
,0566 means 0.566 chances out of 1 that a game where black has played there is 
a win for black.

Could anyone once again confirm that those results sounds correct ?


mean score =2.1726424464558005
79655.5328628921 Playout/sec
Time=12.563621936
Number of playout=1000762.0
Mean moves per sim 111.0673969035

Amaf Result :  (from empty board, black plays first)
||,490||,507||,506||,511||,514||,511||,506||,507||,490
||,507||,524||,533||,539||,541||,539||,533||,524||,507
||,506||,534||,544||,550||,553||,551||,544||,533||,505
||,511||,539||,550||,558||,561||,558||,551||,539||,511
||,513||,541||,553||,562||,566||,560||,552||,541||,513
||,511||,539||,550||,558||,560||,558||,550||,539||,511
||,507||,532||,544||,551||,552||,550||,544||,532||,506
||,507||,523||,534||,538||,541||,538||,532||,523||,507
||,491||,507||,505||,512||,513||,511||,506||,507||,492


PS : how can i do so that my response to this mailing-list will be correctly 
indented ? (for example i would have liked to set this one as a response to my 
previous post).
_
Email envoyé avec Windows Live Hotmail. Dites adieux aux spam et virus, passez 
à Hotmail ! C'est gratuit !
http://www.windowslive.fr/hotmail/default.asp___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] 7.5-komi for 9x9 in Beijing

[Christoph Birk]

On Wed, 8 Oct 2008, Don Dailey wrote:

On Wed, 2008-10-08 at 11:47 -0700, Christoph Birk wrote:

On Wed, 8 Oct 2008, Don Dailey wrote:

much more common.There were just a few games that used 6.5 komi
because when I first started CGOS I had set 6.5 by mistake but I think
that was just for a few hours at most.   The vast majority of these are
7.5 komi games:


After all this discussion about komi for 9x9 games, wouldn't you
think that using 7.5 was a mistake and go back to 6.5 ?


Why?


Because a game where the 2nd player wins would never get started :-)
Assuming 7 is the game-theoretical value.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Light simulation : Characteristic values

[Christoph Birk]

On Tue, 7 Oct 2008, Denis fidaali wrote:

The engine is written in java, and run on a quad core Q9300 @ 2.50 Ghz.
The code has been lightly optimized, and use pseudo-liberties to detect 
captures.


Run it on CGOS, it should get a similar rating to 'myCtest':

name#light_simulations ELO
myCtest-10k 1  1000
myCtest-50k 5  1300

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: OT: Teaching Go (was Re: Disputes under Japanese rules)

[Christoph Birk]

On Thu, 18 Sep 2008, Don Dailey wrote:

It didn't take very long at all before I figured out all the basic cases
for myself.Even the 2 eye rule I had heard of and even understood
it from a book, but it was still rather abstract to me until I actually
experienced it for myself.   Only when it actually happened did the
light bulb go off and I said to myself, this 2 eye thing really is a
big deal.


That bulbs comes on much faster (for most people) if they are
guided towards the light ... like with the methods explained
by others in this thread (eg. capture stones/groups at
corner/side/center)

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Lockless hash table and other parallel search ideas

[Christoph Birk]

On Tue, 9 Sep 2008, Olivier Teytaud wrote:

In 19x19, it's much better, but the MPI parallelization of  9x9 Go is
challenging.


The bright side here is that 9x9 is not really important but just
a test bed. If it works for 19x19, that's good.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Lockless hash table and other parallel search ideas

[Christoph Birk]

On Tue, 9 Sep 2008, Olivier Teytaud wrote:

testbed for
parallelization because it's more difficult) and as real targets (as there
are players
for both).


Sorry, but there are (almost) no players for 9x9. To repeat
D.Fotland's earlier comment: 9x9 is just for beginner's practice.
It's not go.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] cgos 13x13 seems down

[Christoph Birk]

On Fri, 5 Sep 2008, Magnus Persson wrote:
I will also run Valkyria on CGOS 13x13 over the weekend, (or long as things 
are stable).


One anchor would be nice.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] yet a mogo vs human game

[Christoph Birk]

On Aug 25, 2008, at 10:47 PM, Olivier Teytaud wrote:
Just for information, mogo will play in a few minutes (on Kgs /  
computer-go) some games

against high level humans.


MogoTitan is playing 9x9 against nutngo ?

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: What's happening at the European Go Congress?

[Christoph Birk]

On Mon, 11 Aug 2008, Don Dailey wrote:

But let's not exaggerate.   This was not just a simple matter of filling
empty points.


It was.


It was obviously unclear enough to some of us that it required some
analysis.   Even the strong Leela did not see this as merely filling in
the empty points.


That's because it involves a Seki that Leela does not handle properly,
but any 10 kyu should recognize.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: Strength of Monte-Carlo w/ UCT...

[Christoph Birk]

On Aug 10, 2008, at 1:46 PM, Robert Waite wrote:
Exhaustive search is scalable in that I could give it all the  
memory and time it wanted. And it would approach a finite amount of  
memory and a finite amount of time.


Yes, but exhausitve search does not improve your player by 63% (eg.)
for a doubling in CPU time.
This part was done in an empirical scalability study. Please check the
archives of the list.

In the (inifinite) limit minimax+evaluation-function would find the  
perfect move

too, but UCT/MC already find good moves before the limit.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Strength of Monte-Carlo w/ UCT...

[Christoph Birk]

On Aug 9, 2008, at 6:01 PM, Don Dailey wrote:

On Sun, 2008-08-10 at 01:59 +0200, Vincent Diepeveen wrote:

On Aug 9, 2008, at 9:45 PM, Don Dailey wrote:

I'm curious what you guys think about the scalability of monte  
carlo

with UCT.

The MCTS technique appears to be extremely scalable.  The  
theoretical

papers about it claim that it scales up to perfect play in theory.



We agree here that this is not true of course.


No, I think we disagree this time my friend!

Monte Carlo of course by itself is not scalable.  But when combined  
with
tree search such as UCT,  it is equivalent  to a mini-max search  
with a

high quality evaluation at leaf nodes.   It's scalable because the
longer it searches, the more it acts like a proper mini-max search.


Well said, Don.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

[computer-go] Gnugo-3.7.10-a3

[Christoph Birk]

Achor 'Gnugo-3.7.10-a3' loses a lot on time.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] komi for 13x13 and 19x19

[Christoph Birk]

On Aug 2, 2008, at 10:34 AM, Don Dailey wrote:

Does it make sense to use a komi of 7.5 for 13x13 and 19x19 under CGOS
rules?


I don't know about 13x13, but for 19x19 you should use 6.5.

Christoph


___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] 13x13 server up and running

[Christoph Birk]

On Aug 2, 2008, at 1:48 PM, Don Dailey wrote:


Ok,  the 13x13 server is up and running.   Here are some temporary
instructions that will probably be understandable for those with bots
already running:



would be nice to get a few bots on 13x13 to get it started off.


myCtest-10k-UCT is running ...

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] 13x13 server up and running

[Christoph Birk]

On Aug 2, 2008, at 2:23 PM, Christoph Birk wrote:

would be nice to get a few bots on 13x13 to get it started off.


myCtest-10k-UCT is running ...


Weired. I got disconnected during my first game (12) but CGOS
does not mention this game as a loss for myCtest ... it ignored it
entirely and the website does not show game-12.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] linux and windows

[Christoph Birk]

On Thu, 17 Jul 2008, David Fotland wrote:

Not trolling for flames, just expressing an opinion.  If someone is not
willing to put in one day effort to port from Linux to Windows, why should
they expect anyone else to put in one day effort to make Linux available as
a platform?  It seems Linux people are just as chauvinistic as Windows
people :)


Because it's 1 (?) days work for EVERYBODY who has to port, and
not just for one person at the event.
Also, not everybody has a windows computer ...

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: linux and windows

[Christoph Birk]

On Thu, 17 Jul 2008, Dave Dyer wrote:

If your program has ANY gui at all though, you're pretty much screwed.
Mac Windows and Linux GUIs are about as far apart as any three platforms
can be.  There are lots of compatibility solutions, including your
choice of platform independent languages; but they all create essentially
a fourth platform that you have to target, and once again, you're screwed
unless you started that way.


Not quite true anymore; Macs support X-windows, so you can use the
same GUI for Mac+(L)inux.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] 10k UCT bots

[Christoph Birk]

On May 13, 2008, at 7:25 AM, Jason House wrote:
I'm testing my bot on CGOS using pure UCT, no pondering, and 10,000  
playouts per move. Can someone put up a comparable bot?




I will re-start 'myCtest-10k-UCT' later today.

Christoph
 
___

computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] 10k UCT bots

[Christoph Birk]

On May 13, 2008, at 10:00 AM, Jason House wrote:
On May 13, 2008, at 12:00 PM, David Fotland [EMAIL PROTECTED] 
games.com wrote:


When you say pure uct, what is the playout policy?  Pure random  
moves except

don't fill one point eyes?


That's exactly what I meant. I'd also assume other stuff like the  
UCB1 formula, no RAVE, and no initial move bias. I'm not opposed to  
other variants to see the effects, but I do want to ensure my  
implementation is correct.


'my-Ctest-10k-UCT' uses none of these, except guiding the search by UCT.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] 10k UCT bots

[Christoph Birk]

On Tue, 13 May 2008, Mark Boon wrote:

If this asymmetry really bothers you, you could very easily fix this by
wrapping the search around. There's no asymmetry in a circle.


That doesn't fix anything.


Why not? The whole argument is about a bias against points towards the end. 
In a circular list there is no 'end'.


No, it was a bias towards moves behind illegal moves.
Those moves are twice as likely to be played than other moves. 
Consider a list with 5 moves:


[Move1] [Move2] [Move3] [Move4] [Move5]

You create a random number between 1 and 5. If Move2 is illegeal
for example, then you will play
 Move1 if random#=1
 Move3 if random#=2 or 3,
 Move4   =4
 Move5   =5

Move3 is twice as likely to be played. Even if you make a circular
list.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] CG'2008 paper: Whole-History Ratings

[Christoph Birk]

On Wed, 9 Apr 2008, Andy wrote:

For example:  Suppose a player's true strength is 1500 for some time, and
then he suddenly improves to 2000.  Both before and after he plays a fixed
number of games per day (say 10).  Show a graph of what each rating
algorithm would think his rating is over time.  Many people complain that
the KGS algorithm does not move fast enough for a case like this.


I think that's a bad example since no player suddenly improves
by 500 ELO points.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] CG'2008 paper: Whole-History Ratings

[Christoph Birk]

On Wed, 9 Apr 2008, terry mcintyre wrote:

How does 500 elo points compare to kyu ranks?

Beginning players do improve by 4-5 ranks in a short
period of time. We don't all start as dan-level
players, alas!


Yes, but short time will still be many games.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] CG'2008 paper: Whole-History Ratings

[Christoph Birk]

On Wed, 9 Apr 2008, Matthew Woodcraft wrote:

Beginning players do improve by 4-5 ranks in a short
period of time. We don't all start as dan-level
players, alas!



Yes, but short time will still be many games.


It might be that most of those games aren't visible to the rating
system.


That might explain why a rating system may have a hard time
to follow.
Bad data in ... bad data out :-)

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] now: operating systems and love, was: Paper for AAAI (David Silver) PDF problem

[Christoph Birk]

On Tue, 8 Apr 2008, steve uurtamo wrote:

There isn't, and this is actually a fortunate thing, yet any way to
use unix without at some point needing to use a command-line
tool.  This is what will keep it out of the hands of consumers for
a long time to come, but I think that it's an inherent fact of a
secure operating system.


That's why Mac is the best :-)
You have the nice GUI stuff, if you want you can use an xterm
and you can run 'Parallels' (Parallels.com) ie. Windows at the
same time.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] State of the art of pattern matching

[Christoph Birk]

On Mar 31, 2008, at 10:48 AM, Mark Boon wrote:
I don't know about this. I'm pretty sure MoGo checks if the stone  
can make at least two liberties (ladder problem) in which case it  
can still be horrible but very seldomly worse than random.


I would expect playing a not-working ladder to be worse than random  
most

of the time.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] State of the art of pattern matching

[Christoph Birk]

On Mar 31, 2008, at 1:05 PM, Don Dailey wrote:



Christoph Birk wrote:


On Mar 31, 2008, at 10:48 AM, Mark Boon wrote:
I don't know about this. I'm pretty sure MoGo checks if the stone  
can

make at least two liberties (ladder problem) in which case it can
still be horrible but very seldomly worse than random.


I would expect playing a not-working ladder to be worse than random
most
of the time.

Of course this is true, but presumably a move that answers a direct
atari threat would classify as being better than random.


Not if it's a working ladder.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] 9x9

[Christoph Birk]

On Mar 26, 2008, at 12:32 AM, Olivier Teytaud wrote:
 ... is room for improvement. But 19x19 is something else, perhaps we


can have the Dan, but I'm not sure of that in spite of the gentle  
words of

Catalin, and I'm sure the current
mogo can't win against a professionnal player in 19x19 whenever we
have the best cluster in the world, and whenever the professionnal  
player

is both ill and a bit drunk :-)


By reaching 1-dan (amateur) you would have received
1 M$ a few year ago from Mr. Ing.

Christoph



___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

[computer-go] Ing Challenge

[Christoph Birk]

On Mar 26, 2008, at 9:47 AM, David Fotland wrote:
The lower level prizes were given for games against Insei, but the  
top prize

was for play against t top professional.

http://www.smart-games.com/worldcompgo.html

I can't find any official data on-line, but the information in the  
page

above was copied from the paper rules at the competition.



You are right.
The AGA website writes strong amateur.
 http://www.usgo.org/ingfoundation/

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

[computer-go] 9x9 CGOS

[Christoph Birk]

 http://cgos.boardspace.net/9x9/standings.html
is not updating.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Optimal explore rates for plain UCT

[Christoph Birk]

On Thu, 13 Mar 2008, Petr Baudis wrote:

So I have created this page:

http://senseis.xmp.net/?CGOSBasicUCTBots

and summed up what I could find in the thread about the various bots.
Please clarify if anything there is wrong / unknown, and add your bots
if they aren't there. I wanted to add Fluke too, but I do not know which
of the many incarnations should I choose. :-)


I am not sure if we have an understanding of node expansion. myCtest
does not really the parent node ... let me explain what I am doing:

During decending (root at the top) the UCT tree:
 if current-node is a leaf
   if number of visits is at least MIN_VISITS then
 determine all legal moves and create children nodes
 choose a random child and descend
   endif
   run a random playout and propagate score upwards
 else
   calculate UCT score = win-ratio + C * sqrt(log(n)/m)
   decend to best child
 endif


Curiously, while pachi1 with 10k playouts is 30 ELO weaker than
drdGeneric-10k and myCtest-10k-UCT (it seems like ~1230 is _the_ rating
for 10k UCT), with 50k playouts it is 60 ELO stronger than
myCtest-V-0003 - is that one really just UCT with 50k playouts?


Name   #playoutsC  MIN_VISITS ELO
myCtest-10k-UCT:   10k  0.550 1228

myCtest-V-0020:50k, 0.5MIN=50 1459
21:50k, 0.5MIN=25 1483
22:50k, 0.5MIN=10 1467
23:50k, 0.5MIN=5  1523
24:50k, 0.5MIN=2  ?

My explanation is that with fewer playouts the reduced noise with
a larger MIN_VISITS is better, while with more playouts the
deeper search-tree with a smaller MIN_VISITS improves play.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Optimal explore rates for plain UCT

[Christoph Birk]

On Thu, 13 Mar 2008, Heikki Levanto wrote:

Would it make sense to have a similar page for pure MC programs (without
uct), so that we beginning developers could check that portion of our code
against known results?


I have two long-term CGOS programs:

myCtest-10k:  1011 ELO
myCtest-50k:  1343 ELO

Pure MC random playouts, no tree.

There appears to have been a small drift over the last year. Both
programs lost about 40 ELO ponts in the last 400 games.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Optimal explore rates for plain UCT

[Christoph Birk]

On Wed, 12 Mar 2008, Don Dailey wrote:

1.  My UCT constant is 1.0  - my formula is  averageScore + c * sqrt(
(2.0 * log(n)) / (10.0 * m) );


so your contstant is 2/10 = 0.2 inside the sqrt(), which is
equivalent to c=0.44 ?

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Optimal explore rates for plain UCT

[Christoph Birk]

On Tue, 11 Mar 2008, Don Dailey wrote:

If it is agreed,  I will start a 25k test.My prediction is that this
will finish around 1600 ELO on CGOS.


I have long term rating for simple random playouts:

myCtest-10k and myCtest-50k.

I keep them active since Sept/2006. Please don't use 25k.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Optimal explore rates for plain UCT

[Christoph Birk]

On Tue, 11 Mar 2008, Don Dailey wrote:

This isn't simple random play-outs.It's monte carlo with UCT tree
search.

Ok,  I will use 50k to match your test.It means I  probably cannot
run 2 tests on that machine and is why I hoped it would be minimal
resource usage,  but since you have already started I will restart my test.


You can also use 10k as Jason suggested.

myCtest-10k-UCT uses 10k (random/light ) playouts with UCT.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Optimal explore rates for plain UCT

[Christoph Birk]

On Tue, 11 Mar 2008, Don Dailey wrote:

I am going to keep the 25k playouts running and add a 10k play-out
version of UCT. I want to establish a standard testing size so that


Great! That way Jason can also participate.

myCtest-10k-UCT has a long-term rating of about 1250.
For the 50k version I have just started a test series that experiments
with various thresholds before creating a new node.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])

[Christoph Birk]

On Mon, 10 Mar 2008, Petr Baudis wrote:

MoGo displays the depth of the principle variation in the stderr stream.


I have been wondering, does that include _any_ nodes, or only these
above certain number of playouts?  What is the playout threshold?


The 'principal variation' is usually the one that the program would
play against itself; at each level the one move with the highest
score with might (depending on the program) just be the one with
the most playouts.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Optimal explore rates for plain UCT

[Christoph Birk]

On Mon, 10 Mar 2008, Petr Baudis wrote:

 With 110k playouts per move and no domain knowledge in the playouts,
the ratings are now:

c=0.2  (pachi1-p0.2-light)  ELO 1627 (285 games)
c=1.0  (pachi1-p1.0-light)  ELO 1590 (120 games)
c=0.05 (pachi1-p0.05-light) ELO 1531 (286 games)
c=2.0  (pachi1-p2.0-light)  ELO 1511 (118 games)


I have two light UCT bots on CGOS:
Name  #playouts c (*) CGOS-ELO
myCtest-V-00035 0.25  1508
myCtest-10k-UCT   1 0.25  1246

(*): I use c=0.5 outside the sqrt()

What is your 'create-new-node' threshold? I use 50.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Optimal explore rates for plain UCT

[Christoph Birk]

On Tue, 11 Mar 2008, Petr Baudis wrote:

On Mon, Mar 10, 2008 at 06:57:07PM -0400, Don Dailey wrote:

I think you may still have a bug.  You should get well over 1700 with
110,000 playouts, even if they are light playouts.


I will run myCtest with 110k-playout, c=0.25 and node creation
after the 2nd visit ... let's see what its ELO rating will be
in a couple of days.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Optimal explore rates for plain UCT

[Christoph Birk]

On Mon, 10 Mar 2008, Christoph Birk wrote:

On Tue, 11 Mar 2008, Petr Baudis wrote:

On Mon, Mar 10, 2008 at 06:57:07PM -0400, Don Dailey wrote:

I think you may still have a bug.  You should get well over 1700 with
110,000 playouts, even if they are light playouts.


I will run myCtest with 110k-playout, c=0.25 and node creation
after the 2nd visit ... let's see what its ELO rating will be
in a couple of days.


Sorry, I just realized I cannot do 110k playouts because my
implementation is too slow.
I suggest you run a 'pachi-0.25-light-50k' that just uses
5 playouts. That way you can compare it to 'myCtest-V-0003'.

BTW: I count the new-node threshold like Don from the parent
 node, so 50 not far from your '2'.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Floating komi

[Christoph Birk]

On Thu, 6 Mar 2008, Don Dailey wrote:

One last time: Nobody suggested a one fix for all positions/problems.
The floating komi was suggested to guide the UCT search along
certain lines of play during specific (close!) endgame positions.

When I said all positions I meant all games.You expect to apply this
to all winning and losing positions in every game, not just specific ones.


No.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])

[Christoph Birk]

On Thu, 6 Mar 2008, Don Dailey wrote:

And can I assume the tree portion is also inhibited from seeing this due
to a combination of factors such as heuristics to delay exploring ugly
moves as well as  the weakness of the play-outs in this regard (which
would cause the tree to not be inclined to get close enough to the issue
to understand it properly?)


It might also be reading-depth. Some nakade forms need quite deep reading
you want to discover them on-the-fly. 10 kyu humans know that the
bulky-five is dead; no reading required.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])

[Christoph Birk]

On Thu, 6 Mar 2008, Don Dailey wrote:

advantageous to give away stones that not.  Despite what many people
believe,  MC programs don't normally believe it's better to win small
and they are not hell-bent on giving away stones in order to try to make
the score come out to be exactly 0.5 win.


You are correct that it is not explicitly programmed into the MC
programs to win by 0.5 pts, but since most of them don't care about
the margin they in practice often do.
As you might remember I (3 kyu) played many games on CGOS and many
games that I lost, I actually lost by 0.5 pts. It mostly worked like
this: I am behind by several points in the early endgame, then the
programs allow me to gain a point here, or there. But in the end
they still win by 0.5 pts.
These programs are NOT hell-bent on losing, they just dont care
if the UCT-tree shows that they win anyway.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])

[Christoph Birk]

On Thu, 6 Mar 2008, Weston Markham wrote:

You are right, but I think that you may also be misconstruing the
nakade problem as a lack of concern about margin, when it is really a
fundamental failure to understand (i.e., failure to explore


Sorry, you miss-understood.
The nakade problem is totally unrelated to the margin problem.
They just sometimes happen at the same time and then allow
someone to take advantage of them.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Floating komi

[Christoph Birk]

On Mar 5, 2008, at 11:58 AM, Don Dailey wrote:

Don Dailey wrote:

 not assuming that MC plays the best move.   The problem isn't the

assumptions I am making, but the assumptions others are making,  that
it's NOT playing the best move.You want to apply a fix to all
positions without really knowing which positions are a problem.


One last time: Nobody suggested a one fix for all positions/problems.
The floating komi was suggested to guide the UCT search along
certain lines of play during specific (close!) endgame positions.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: Should 9x9 komi be 8.0 ?]

[Christoph Birk]

On Tue, 4 Mar 2008, Magnus Persson wrote:
But here you are missing the point that close to 0% winning probability means 
that it cannot win against random play. The opponent  could lose only by 
killing his own groups.


I don't know why you (and Don) keep bringing up the 0% against random
play ...
I am talking about a (typical) situation in the endgame
where best play (as seen from the program) leads to a sure 0.5 pt loss.
Many MC programs will make unreasonable attempts of winning by chosing
a line that shows a possible win (10 pt) if the opponent makes a
(stupid) mistake. Instead they should go for the (supposedly sure)
0.5 pt loss, because the opponent will much more likely make
the 1pt mistake, and not the 10 pt mistake.

The problem is that the likelihood of your opponent making a mistake
is hard to determine by the UCT (MC) playouts. I guess one needs
to use  the meta information that is is more likely to make a small
mistake than to make a big one.

This is not specific to any particular opponent.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: Should 9x9 komi be 8.0 ?]

[Christoph Birk]

On Tue, 4 Mar 2008, Don Dailey wrote:

I really believe the source of peoples confusion on this is believing
that the program starts playing ugly random moves as soon as it is
down a little. But in fact, when it gets into ugly mode it is
because the score is very close to 0.0  or in some programs like Lazarus
-1.0


I thinks it's the source of your confusion. A MC program makes
silly moves when it thinks it will definitely lose; but it
does so even if the margin of loss is only 0.5 pts.

You are right, if the margin is 20 pts, the game is lost anyway
and one should better resign. We are talking about a small (by score!),
but sure (as estimated by the program) loss.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: Should 9x9 komi be 8.0 ?]

[Christoph Birk]

On Tue, 4 Mar 2008, Don Dailey wrote:

When you get into opponent modeling,  you have to understand your
opponent, because usually opponent modeling involves playing weaker
moves in exchange for better practical winning chances.


No, I don't want to do any opponent modelling.
And no, opponent modelling does usually not involves playing weaker ...


If that's really what you want,  why not just using the territory
scoring method instead of the win/loss record for your MC player?


That is not what I am suggesting. I am just talking about specific
situations in the endgame.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: Should 9x9 komi be 8.0 ?]

[Christoph Birk]

On Tue, 4 Mar 2008, Magnus Persson wrote:
I do not see why an MC programs in general is biased towards winning with 10p 
instead of a single 1p mistake.


It is not biased, that's my point.
It should be biased toward the '1pt' loss, if loss is unavoidable,
not for beauty but for the likelihood of converting a 1pt loss into a win.

I am getting tired too, and I is a detail, not (yet) very important.
But I thought it might be a relatively easy one to fix using the
proposed situation-dependent komi, and small improvements tend
to add-up over time :-)

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: Should 9x9 komi be 8.0 ?]

[Christoph Birk]

On Tue, 4 Mar 2008, Weston Markham wrote:

greater loss by the program.  (You also characterize the opponent's
blunder in (b) as stupid, but I understand this to simply be a
subjective characterization based on the fact that it leads to a large
loss.)


In my own experience it is much easier to make a 1pt mistake
than a 10pt mistake. This may be 'subjective' but it's still true.


(b), but it is sure to die as long as white responds at (a).  The rest
of the board is such that black will lose by 0.5 anyway if he plays
the normal move at a, and white responds with (b).  Do MC-based
programs really favor (b) over (a)?  (I am skeptical.)


No, they don't.


Or is the
issue that Christoph and others believe that players should (for
whatever reason) favor (a)?


Yes ... but not for 'whatever' reason :-)

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: Should 9x9 komi be 8.0 ?]

[Christoph Birk]

On Sun, 2 Mar 2008, Don Dailey wrote:

My feeling is that in lost positions,  the only thing we are trying to
accomplish is to make the moves more cosmetically appealing (normal) and
at best improve the programs chances of winning against weak players.
After all, if the program is in bad shape,   then to be completely
realistic it's probably going to lose to the player that put it in this
bad shape.


I think you are wrong here.
If there are two lines of play from the viewpoint of the MC program:
 a) leads to a 0.5 pt loss
 b) may win if the opponent makes a stupid (!) mistake, but otherwise
leads to a bigger loss.

It is generally better to play for the 0.5 point loss as the oppoenent
may make a end-game mistake and loses 1 point.
But naive MC programs typically go for (b) which will lead to a
devastating loss because the opponent usually does not make the 10 point
mistake, but may have made the 1 point mistake.

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: Should 9x9 komi be 8.0 ?]

[Christoph Birk]

On Sun, 2 Mar 2008, Don Dailey wrote:

This is true in GO too.   I'm talking about the kinds of position where
go program start to play aimlessly and they only do that when the
result is like being down a queen in chess.Even being down a piece
in chess is playable if there is some compensation.


No, it's not a 'queen'. MC Go programs start playing aimlessly even
if 0.5 points down, if they are sure about it.
It would be much better (looking AND for winning) to nevertheless
follow that 0.5 point losing line and hoping the opponent makes
a 1 pt. mistake instead of trying useless and obviously failing
invasions.

Christoph

___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Re: Should 9x9 komi be 8.0 ?]

[Christoph Birk]

On Mon, 3 Mar 2008, Don Dailey wrote:

What you are trying to do is more in the category of opponent
modeling.You want to optimize for the case that you might
occasionally salvage a game against an opponent that is much weaker than
you but is beating you anyway.


No, absolutely not. The idea of following the 0.5 pt loss is always
true, even if the opponent is of comparable strength.


strength level.  If your program KNOWS it is losing by 0.5 points,  then
it's reasonable to expect that your opponent does too, especially given
the fact that he just outplayed you.


I think you are too much of chess player :-)
The fact that he is 0.5  point in the lead does not imply he is
(much) stronger. Any player, in particular a human player, is capable
of the making a mistake. So it is important to stay on the 'small'
losing line. That might a difference to chess, where there is no
'small' loss.


So at best you hope your opponent will make a stupid mistake in an
obviously lost position for you.


No, the opposite. Not a stupid mistake; I am hoping for the subtle
mistake. But you throw that opportunity away If you play desparate
moves just because you think you will lose the game by 0.5 points.


There is nothing wrong with this,  if it's what you want to lose sleep
over,  but how much do you expect to gain from it? I see people
getting excited about this idea as if it's the holy grail of computer go
and will add 50 ELO or more.


Nobody called this the holy grail ... but I agree with you that
there  are bigger problems in computer Go ...

Christoph
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/