Petr Baudis | 27 Nov 15:37 2015

[66,866 human-pachi games]


  Would anyone find these useful?

----- Forwarded message from Jonathan Chetwynd -----

I recently took pachi/ offline due to excessive use, mostly from

a visualisation of the site statistics by country and player is here:

there are 66,866 human-pachi games in database,
if anyone has idea how to use, let me know.


I may update to Pachi 11.0 but this is dependent on new web server.....

----- End forwarded message -----

  There's probably no way to label the games by opponent strength, and
I don't know if these are necessarily complete games.  But they are all
played against a fixed opponent, MCTS program.

				Petr Baudis
Computer-go mailing list
Computer-go <at>
(Continue reading)


CFP:: DICTAP2016- IEEE- Turkey

The Sixth International Conference on Digital Information & Communication
Technology & its Applications (DICTAP2016)

Mevlana University, Konya, Turkey

July 21 - 23, 2016.

The conference aims to enable researchers build connections between
different digital applications.

The conference welcomes papers on the following (but not limited to)
research topics:

- Information Content Security
- Web Services, Web based Application
- Scientific Computing and Multimedia Processing
- Sensor Networks and Embedded System
- Transmission, Antenna & Propagation
- E-Learning, E-Commerce, E-Business and E-Government

Important Dates:

Submission Deadline       : Open from now until June 21, 2016
Notification of Acceptance: June 30, 2016 or 4-7 weeks from the submission

(Continue reading)

Clark B. Wierda | 26 Nov 02:53 2015

Michi-go: Version 1.0 released

I've released version 1.0 of michi-go.  The code is on Github at

This version plays about twice as fast as the Python code on PyPy, but only has a winrate of 24% for an ELO difference of -200.

I'm not sure where the weakness is coming from, but I intend to investigate the Random Number Generator, the RAVE calculation, and try to find any logic errors introduced by the translation.

This version lines up fairly well with the original Python.  Like the original, it is all in a single source file.  The command line options are handled the same and all function.  Currently, I am using the same 1400 playouts as the Python code.

I will maintain the version_1 branch as matching the structure of the original Python.  If any issues are found in the original translation, I will update that branch.

Version 2 will work to be more consistent with Go idiom.  I plan to add the ability to play any size board (at least 9, 13, and 19), handle handicaps, and provide the option to use positional superko.  I will also investigate what changes are needed to match michi-py in strength.

Clark B. Wierda
Computer-go mailing list
Computer-go <at>
Yuandong Tian | 24 Nov 21:45 2015

Re: Facebook Go AI.

Hi all,

I am the first author of Facebook Go AI. Thanks for your interest! This is the first time I post a message here, so please forgive me if I mess up with anything.

1. The estimation of 1d-2d is based on the win rate of free game in the last 3 months (since darkforest launched in Aug). See Table 6 in the paper. For ranked game, its rank is definitely lower since people tend to play more seriously. It seems that now darkforest is 1k and darkfores1 is 1d. 

2. Here is the Pachi 10k command line for no pondering.
pachi -t =10000 threads=8,pondering=0

For pondering, it is simply 
pachi -t =10000 threads=8

In both cases, all the spatial patterns are properly loaded. See the following GTP response:
W>> protocol_version
Random seed: 1448000132
Loaded spatial dictionary of 1064482 patterns.
Loaded 3021829 pattern-probability pairs.

3. We use pachi version 11.99 as shown in the following GTP response:
W>> version
W<< = 11.99 (Genjo-devel): If you believe you have won but I am still playing, please help me understand by capturing all dead stones. Anyone can send me 'winrate' in private chat to get my assessment of the position.  Have a nice game!

4. Darkfores2 is still DCNN model and no search is involved.

Thanks! If you have any comments, please let me know.

Yuandong Tian
Research Scientist,
Facebook Artificial Intelligence Research (FAIR)
Computer-go mailing list
Computer-go <at>
Harald Korneliussen | 24 Nov 20:15 2015

Why do Facebook use DNN next-move predictors? Shouldn't they use next-move generators?

When I read about Facebook's DCNN-using go program, I remembered another paper that I'd come across on arxiv, namely "How (not) to train your generative model: scheduled sampling, likelihood, adversary?" by Ferenc Huszar ( 

A lot of that paper went over my head (I am a "half-studied scoundrel" as we say in Norway), but his speculation in the end, I think I sort of got, and it made a lot of sense to me.

He argues that which side you approach the K-L divergence from so to say matters for what kind of errors you get when the model, and that when you're generating as opposed to predicting, the goal should be to minimize the K-L divergence from the "other" way.

When you're using a DCNN in a go program, you are really doing generating, not prediction, right? You want to generate a good move. A model that generates "flashy" moves that LOOK really strong, but could potentially be very bad, would be a good predictor, but a bad generator.

The ideal probability distribution is the distribution of moves a pro would make. But to the degree your model falls short, you want to minimize the chance of making a wildly "un-pro" move, rather than maximizing the chance of making a "pro" move. Since these are probability distributions, those two things are not the same unless your model is perfect (right?).

If my understanding is correct (and it's quite possible I'm way off course, I'm an amateur! sorry for wasting your time if so!), then rather than training a move predictor, they should use the adversarial methods which are also in the wind now to train a generative model. 

-- Harald Korneliussen

Computer-go mailing list
Computer-go <at>
Nick Wedd | 24 Nov 19:10 2015

November KGS bot tournament: 19x19, and December KGS bot tournament: 13x13

Thank you, Hideki, for pointing out my error. I hope the title of this email helps to correct it.

Registration is now open for two KGS bot tournaments:

November 29      19x19, 14 minutes each, starts at 08:00 UTC, 12 rounds
December 06      13x13,   9 minutes each, starts at 16:00 UTC, 18 rounds

Nick Wedd      maproom <at>
Computer-go mailing list
Computer-go <at>
Nick Wedd | 24 Nov 18:10 2015

December KGS bot tournament: 19x19

The December KGS bot tournament will be on Sunday, December 6th, starting at 16:00 UTC and ending by 22:00 UTC.  It will use 13x13 boards, with time limits of 9 minutes each plus fast Canadian overtime, and komi of 7.5.

Please register by emailing me, with the words "KGS Tournament Registration" in the email title, at maproom <at> .

Note that the November KGS bot tournament (November 29th, 19x19, 08:00 UTC start, 14 minutes each) is this coming Sunday, very late in the month because there have been other computer Go events in Japan, Beijing and Taiwan on the last three weekends. The December KGS bot tournament is the following weekend.  If you email me to register for just one of these events, please make it clear which one you mean.  If you email me to register for both, please make that clear too.

Nick Wedd      maproom <at>
Computer-go mailing list
Computer-go <at>
Horace Ho | 24 Nov 07:09 2015

Looking for game record of World Computer Weiqi Championship in Beijing

I am looking for the game records between Lian Xiao and DolBaram in Nov 2015.

3rd game is here:

How about the firs two games?


Computer-go mailing list
Computer-go <at>
甲斐徳本 | 20 Nov 15:45 2015

Demis Hassabis (DeepMind) on Go

DeepMind may be coming to Go:


Computer-go mailing list
Computer-go <at>
Marc Landgraf | 19 Nov 16:13 2015

Theoretical question

there is a question that lately crossed my mind. Considering an nxn Go board, no suicide allowed and with a rule that does not allow repetition of a position, unless caused by a single pass:
What is the maximum number of board positions that can be run through in a single sequence starting from an empty board?
Different wording: What is the maximum game length if not counting passes?

It can easily be proven that for n>=3 this number must be lower than the number of legal positions. I didn't check for n=2, simple bruteforce would probably solve that.

What happens to the ratio between the number of legal positions and the longest possible sequence with growing n?

Is this already known? Or does anyone have a clue how to figure it out?


Computer-go mailing list
Computer-go <at>
Joshua Shriver | 19 Nov 14:46 2015

CGOS boardspace

I did a restart of the 9x9 and 19x19 as a test. Anyone mind testing it
to see if you can connect?
I left 13x13 alone since it just worked.

Know a couple people have forked the cgos codebase which is wonderful!
 If anyone needs help Im willing to share info, or help best I can.  I
just had such a hard time getting the original version to stay up and
do apologize.

If someone would like to take over the cgos I can ask the site owner
to make sure then arrange a transfer of login credentials.

I sadly don't have the time to dedicate to it anymore.


P.S. Sorry I let Don down.
Computer-go mailing list
Computer-go <at>