Codecentric Challenge 2016: Lukas Krämer (6d) vs. Zen

[Bonobo]

CODECENTRIC CHALLENGE 2016: Lukas Krämer​ (6d) vs. Zen

“Best-of-5” Match
Lukas was German champion 2013, 2014, 2015.
Zen won gold medals on 19x19 in Computer Olympiads 2011, 2013, 2015, 2016.

The games will always be on a Sunday evening (for Central Europe), beginning at 18:00 CEST, in the “computer room” on KGS.

First game is on Sunday, August 28.

http://computer-go.org/pipermail/comput ... 9507.html

[pookpooi]

byoyomi this year is 20 secs longer than in 2014-2015
Lukas and former human FJ Dickhut's rating are roughly the same (Lukas is a little bit stronger)
Lukas is also younger, and has been playing with bots for a while (his KGS account miao 6d) so he might be more suitable for Zen
I'll watch it closely, as I think this will be the last codecentric challenge.

[pookpooi]

just a little update, zen is online with new version and won 12 out of 13 games against cloudnotes 8d in even game, 15 seconds 10 byoyomi setting (many suspect that cloudnotes is andy liu (american pro) cause he's also using cloudnotes name in IGS server) and push cloudnotes back to 7d. Look like a tough match for Lukas for sure, Lukas is witness the game but he has no comment. Hideki (co-zen's programmer) also comes in the room when zen play and won against winy1 (another 8d account) and said 'this is just the beginning...'

[Uberdude]

In one of those game gghideki (#2 Zen programmer) said that the latest Zen version 12.1 has a value network. It's probably not as good as AlphaGo's yet, but I didn't expect other teams to make even a basic one this soon, so that's a remarkable achievement.

[johnsmith]

So, version 12.1 uses GPU, but 12.0 doesn't. How much does this GPU improve Zen's strength and is Zen now using something similar to Alphago value network?

[Cassandra]

This version at KGS is Zen 19K (result of the project that is sponsored by Dwango).

Lukas will play against the "older" version Zen 19X.

[pookpooi]

This version at KGS is Zen 19K (result of the project that is sponsored by Dwango).

Lukas will play against the "older" version Zen 19X.

this is news to me
but the older version is also very strong, probably 7d on long time setting

[ez4u]

This version at KGS is Zen 19K (result of the project that is sponsored by Dwango).

Lukas will play against the "older" version Zen 19X.

this is news to me
but the older version is also very strong, probably 7d on long time setting

AFAIK the different Zen accounts are all continuously updated to the latest engine. Zen19x was running 12.0 the last time it played a game on the 23rd. The 12.1 version appeared on the 27th in Zen19k. You can see that info at the beginning of each sgf file in the KGS archives.

Edit: From a comment by the programmer during game 1 it seems that the Zen19x network does not have the right GPUs to run version 12.1

[GoldenBear]

On the codecentric blog you can read a pre-match interview with Lukas Krämer and Hideki Kato: https://blog.codecentric.de/en/2016/08/ ... enge-2016/

BTW: The official website for the match is https://go.codecentric.de

[Bonobo]

1st game begins in ten minutes … —> KGS —> Computer Room

After the game, you will be able to download the SGF here: http://www.gokgs.com/gameArchives.jsp?user=zen19x

<edit>

see also https://go.codecentric.de/2016-game-1/

</edit>

[pookpooi]

lukas won the first game, another evident that human use time more efficiently than bot
so zen19x is 6d on long time setting (8d with blitz magic!)

[Uberdude]

In my and most observers opinion Zen was winning for the opening and most of the middlegame, but played a bit overly safely and then maybe a dodgy ko threat and Lucas reversed the game near the end. Yesterday I was at a tournament with Lucas Baker who works at Deepmind and he said AlphaGo thought Zen was winning until around move 150-something (isn't he lucky to get to ask AlphaGo questions like this!).

[pookpooi]

I was joking in the match room that Aja Huang might surprise us with AlphaGo winrate for this match LOL
Anyway, look like Zen has life & death problem that it's realized later in move 200+ which is already too late.

[johnsmith]

In my and most observers opinion Zen was winning for the opening and most of the middlegame, but played a bit overly safely and then maybe a dodgy ko threat and Lucas reversed the game near the end. Yesterday I was at a tournament with Lucas Baker who works at Deepmind and he said AlphaGo thought Zen was winning until around move 150-something (isn't he lucky to get to ask AlphaGo questions like this!).

Wow, thank you for the insight! I wonder how big was the lead AlphaGo had seen?

P.S. Post from Mr. Aja Huang 1 hour ago: "All AlphaGo commentaries have been proof read and are ready to be released once our new DeepMind website is set up. Also glad to see Zen reached 8d on KGS. Looks like with a value network Zen is approaching AlphaGo v13 on the same hardware"

[Krama]

In my and most observers opinion Zen was winning for the opening and most of the middlegame, but played a bit overly safely and then maybe a dodgy ko threat and Lucas reversed the game near the end. Yesterday I was at a tournament with Lucas Baker who works at Deepmind and he said AlphaGo thought Zen was winning until around move 150-something (isn't he lucky to get to ask AlphaGo questions like this!).

Wow, thank you for the insight! I wonder how big was the lead AlphaGo had seen?

P.S. Post from Mr. Aja Huang 1 hour ago: "All AlphaGo commentaries have been proof read and are ready to be released once our new DeepMind website is set up. Also glad to see Zen reached 8d on KGS. Looks like with a value network Zen is approaching AlphaGo v13 on the same hardware"

Deepmind team is lying, I mean how long does it take to setup a website?

They are up to something and I know it.