Lc0 will participate at the WCCC? Wow!

[Martin Renneke]

Hi all. That sounds very interessting. Refer to the blog for details. WCCC is the world champion chip for chess computer. Tomorrow the player will setup there environment and on Friday 13th July they'll start. But who is the player using lc0 and what kind of machine is he using? Thx. Martin. 

Refer here for the WCCC schedule info: http://icga.org/wp-content/uploads/2018/02/ICGA-Chess-Schedule-2018-2.pdf

[John D]

not sure; since it wasn't in the blog i'd imagine nothing was official as of the writing. which presumably means it was just decided by the organizers & how the hell are they gonna do this in 2 days??

[Jesse Jordache]

There's been grumblings about it for a while, championed on the con side by Chris Wittington who's pointed to ethically questionable actions by the group that organizes the WCCC, and on the pro side by no one in particular, arguing that as the TCEC was putting off making a decision about GPU's, presumably while they consulted the pay engines that Leela makes look silly (a subset that does NOT include Stockfish), it's likely that they'll return a CPU only decision which leaves the WCCC as the only available competition.

[M]

Surprising news indeed, given that the tournament will already start tomorrow. But why doesn't Leela participate in blitz?

Last year WCCC was well before the Alpha Zero shock to the computer chess world. It will be interesting to see if the deep learning approach has influenced other participants in some way.

[Matt Blakely]

Where can I follow the games and/or results?

What is the time control?

I can't find much good info on google

[Martin Renneke]

Am Mittwoch, 11. Juli 2018 19:04:29 UTC+2 schrieb Martin Renneke:

Hi all. That sounds very interessting. Refer to the blog for details. WCCC is the world champion chip for chess computer. Tomorrow the player will setup there environment and on Friday 13th July they'll start. But who is the player using lc0 and what kind of machine is he using? Thx. Martin. 

Refer here for the WCCC schedule info: http://icga.org/wp-content/uploads/2018/02/ICGA-Chess-Schedule-2018-2.pdf

From Borg I got the info that this was discussed somewhere in June already in the dev community. A 8 GPU Computer with P40 or newer Nvidia from Western Digital Research will be used in the world championchip. From my understanding a lot of effort was spend to make lc0 working on multi GPU environments. I think this year it's a little bit early for the world championchip, but this might help to get more GPU power in place. Let's look what will happen at the WCCC. 

The rules are here: http://icga.org/?p=2343

It's a 45 minutes game + 15 Seconds per move. So let's keep fingers crossed for the tournament. 

[Dietrich Kappe]

No big discussion. A third party with a ton of gpu offered to do it and it was “pax vobiscum” from there. We’ll see how it goes.

[Nabil Danial]

Cyril (the operator/rep of leela on WCCC) said on Discord that it is going to be 8 V100s. Its gonna be awesome.

[Jesse Jordache]

What kind of processing power do other engines bring to the event?

[Jesse Jordache]

I think a lot of.. not to use the word "tantrum", but a lot of "who entered Leela?  Whooooo did it?  Let me tell you why this is a bad idea.  And here's the same reason again.  The devs need to consider because they're going to lose a ton of credibility.  Well I guess no one here cares about credibility" etc, although at the end the general concept of "open source and its consequences: 101" were brought up.

[Nabil Danial]

Jonny used to use a cluster of 2400 cores in several WCCC events in the past, so I'm expecting the same this time. GridGinkgo was using some 200+ cores cluster and Komodo was on a 60+ cores machine. I don't know what they plan to use this time around though.

[Ron Langeveld]

Sorry Jesse, but is there any coherent thought coming up soon? I don't have the faintest clue what you trying to say here. I hope it doesn't refer to one of your other posts. It's hard to keep track on a train of thought via this forum.

[Matt Blakely]

Care to share why?

Please don't say because you think it will do poorly.  Because on that hardware it should compete for the win.

[Dietrich Kappe]

He’s referring to the Rybka kerfuffle.

[Chris Whittington]

There's also the zxvega plus issue

[FWCC1]

Thought Leela HAD TO USE CPU ONLY

[FWCC1]

Looks like there is a speed chess format also,Leela should do well in that format. NN seem to have a great intuitive grasp of chess.

[Martin Renneke]

I'm sorry. I misunderstood the schedule information. The WCCC - World Championchip for chesscomputer will start at Monday 16th July. 

[Albert Silver]

Why would it need to use CPU only??

[FWCC1]

Just read the rules she can use GPU,they say 8xV100

[Rudolf Posch]

Currently Leela Chess Zero plays live against Komodo, see http://view.livechesscloud.com/7c104c94-c119-441e-a444-b0f22a1880d0

"Live -.pgn" see  http://www.icga.org/live/round-1/games.pgn

[John D]

thanks for the heads up, but i'm not getting anything with either link (second one is a redirect to itself, first has no action, including opponent names

[Tryfon Gavriel]

It is a bit tricky:

Click on the "vs"

Then click on Round 1

[John D]

got it now, thanks guys

[Chris Whittington]

looking like a draw (move 38), although Leela can often manage to snatch a defeat from drawn position

[Tryfon Gavriel]

This seems to be an endgame thriller from what I just saw - I thought Komodo's plan of a4 was to make Nb5 winning - by winning the a3 pawn

But then Leels played the super mysterious Rg8 - and I could only imagine it is to do with f4 and somehow breaking on the kingside - to try and queen the h pawn!

There is thrilling stuff going on behind the scenes here!

[Chris Whittington]

it is a draw, but Leela has a little problem here, she thinks she is +0.75 or something, which means that, rather than accept a draw by repetition, she'll tend to play something else - that might be the blunder.

difficult to see and AB program with 40 ply or so endgame lookahead blundering here

[Chris Whittington]

whereas, in this position, Komodo does not have much in the way of positional guidance and is really dependent on lookahead not positional. Leela the opposite.

[Tryfon Gavriel]

Do they have a normal 50 move draw rule or has it been extended for this competition to something higher like 75 or 100 ?

Also what hardware is Komodo running on? I have noticed this Wiki page, and Komodo is the reigning champion last two years:

https://en.wikipedia.org/wiki/World_Computer_Chess_Championship

[Chris Whittington]

I assume 50 move rule. chess programs are coded for that count.

last pawn move was #52, so not far to go now

[Chris Whittington]

101 h4 and Leela resets the 50 move counter

[Galaga]

125. Draw acc. to Tablebases  https://www.shredderchess.com/online-chess/online-databases/endgame-database.html

[Enqwert]

Are they allowed table base? , this is tb draw. If not white has some chance to win.

[M]

Looks like Leela has lost. What has happened? Timeout?

[Chris Whittington]

Komodo will have EGTBs presumably. Leela has no EGTB but the neural network eval in this P on R-file endgame will act as a perfectly good EGTB. So, consider it a draw, both sides have enough chess "knowledge" at this stage of the game

[daneao....@gmail.com]

it seems Leela lost due to lack of time... too bad.

[Chris Whittington]

Weird. Time loss for Leela? Oh well, Karma strikes again. Leela should not have been entered into the evil tournament. Can't say I didn't warn them.

[Hasan ÇALIŞIR]

Not too bad..Idiotic..What the hell is going on?

16 Tem 2018 Pts 14:42 tarihinde <daneao....@gmail.com> şunu yazdı:

it seems Leela lost due to lack of time... too bad.

--
You received this message because you are subscribed to the Google Groups "LCZero" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lczero+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/lczero/93972e62-4d93-43ea-8db9-3cab4aa70be0%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[M]

Cant follow your last implication. I'd say it looks like there is some timing problem, maybe related to the extra delay of the human operator. That should be easily fixable though.

[Hartmut Hering]

Would not have happened, if the development team had integrated the tablebase functionality before starting at such an event...

[Martin Renneke]

The second round has started: Leela vs. booot. 

you might enter via: https://icga.org/?page_id=2469

or: http://view.livechesscloud.com/7c104c94-c119-441e-a444-b0f22a1880d0

click at "vs" and select board 4. 

[graci...@gmail.com]

Can't the result be simply changed by the arbiter as TB draw? This is an engine's tournament why the hell human errors are a factor to the tournament?!

And kinda bad sport for komodo's part if it didn't even accepted draw on an obvious position, feels like a desperate move to keep their program on top.

[Hartmut Hering]

Well... it is not fair what the komodo-team was doing there. But it is understandable. Becoming "World-Champion" means to earn money with their program. And the customers only see. "Komodo is champion". They don't see how he became champion at last... You cannot expect fair play under those conditions...

[Jesse Jordache]

Any reason you don't just assume you weren't part of the conversation?  You can ask things to be explained to you, and you can be rude, but if you do them at the same time you won't get any help.

On Wednesday, July 11, 2018 at 5:46:31 PM UTC-4, Ron Langeveld wrote:

Sorry Jesse, but is there any coherent thought coming up soon? I don't have the faintest clue what you trying to say here. I hope it doesn't refer to one of your other posts. It's hard to keep track on a train of thought via this forum.

On Wednesday, 11 July 2018 23:16:40 UTC+2, Jesse Jordache wrote:

I think a lot of.. not to use the word "tantrum", but a lot of "who entered Leela?  Whooooo did it?  Let me tell you why this is a bad idea.  And here's the same reason again.  The devs need to consider because they're going to lose a ton of credibility.  Well I guess no one here cares about credibility" etc, although at the end the general concept of "open source and its consequences: 101" were brought up.

On Wednesday, July 11, 2018 at 4:35:41 PM UTC-4, Dietrich Kappe wrote:

No big discussion. A third party with a ton of gpu offered to do it and it was “pax vobiscum” from there. We’ll see how it goes.

[Albert Silver]

Yes, it is Komodo's fault that Leela and the operator lost on time.

[Enqwert]

The operator could sac the rook for the pawn to earn an immediate draw in the first game. Is it allowed?

[Albert Silver]

No, the operator must execute the moves chosen by the engine. He is an intermediary, but this is not a Centaur game.

[georgi stoilov]

Which version is used of leela n what hardware?
Which version is used for komodo and what hardware? Is it monte karlo komodo?
Did leela lost on time really?
Someone answer please, yhank you.

[georgi stoilov]

Leelas hardware is 8xV100 wuut

[Usman Shahid]

At one stage, Komodo had 5 minutes left and Leela had around 15 minutes but then leela wasted so much time.

[Ron Langeveld]

LOL

You dive in practically each and every discussion here uninvited as if you are a moderator or super expert of some sort, and when on such an occasion you talk 'gibberish' you feel offended when you get a provocative reply. 'Keep up the good work' <\sarcasm>

...and I wasn't asking for help, especially not from you!

[Martin Renneke]

What's going on in the first game. Komodo winning unfair?

http://talkchess.com/forum3/viewtopic.php?f=2&t=67949

From my understanding they are using a Fischer Clock. Because of that 15 seconds will come on top per move and the Komodo Operator was very quick. He has increased the time from 4:39 to 7:13 during the last 30 moves. In move 119 it was according to endgame books already a draw. I would assume that Leela moved very quick for the last moves, but according to the discussion in talkchess @cyril was not able to move within 15 seconds all the time. All what was required is to reach move 170 and than nothing more to do, because than the 50 move rule will stop the trouble. 

An other option might be, that Leela moving to slow in the endgame, but I don't think so. 

[Hartmut Hering]

Well... I saw the game live on the server. The truth is somewere netween... Leela really moved a little bit slow while Komodo answered very quickly. But... Komodo had Tablebases and it was a 6-piece-draw. And they were not able to give lc0 the ability to use tablebases before the WCCC began. So she tried to avoid the 50-move-rule in Move 100 moving the h-pawn. I guess without the tablebase-knowledge she saw a smart advantage for herself. It was not really fair from the Komodo-team to decline the draw offer, but of course they had the right to decline. I don't know what GUI the operator used, but he should give himself a little bit more operator time... It will still not be impossible for Leela to win but it will be much more difficult now after this unnessessary loss.

[Martin Renneke]

Yes. In the chat it was mentioned, that the time for the operator to move was 5 of 15 seconds. In the second game he changed that. Is it already clear that Leela will get tablebases implemendet? I thought she will learn the game without external help. 

[Hartmut Hering]

To your question: "Is it already clear that Leela will get tablebases implemendet? I thought she will learn the game without external help."

Well, of course I don't know the plans of the developers, but in my understanding there is a difference between the phase of "learning" and the phase of "playing in a tournament". Of course. For learning we can discuss a lot about tablebases yes or no. There might be reasons for the usage of tablebases and also reasons against. But for a tournament Leela can train as much as she wants. She will always have a disadvantage if she doesn't recognice a deadly drawn position. Therefore I don't see why tablebases shouldn't be implemented for practical play. By the way: Tablebases are implemented in LCZero 0.10. But they are not implemented in the newer LC0. And I really wonder why.

[Chris Whittington]

There's been a good deal of discussion that Leela has no EGTB (unlike Komodo of course) and that therefore it was fine to play on. If both programs knew it was a draw, it should have been declared a draw, or at least the operators should have agreed draw or accepted a draw offer. So said Lefler, no less.

Lefler, Talkchess: "You might ask why we played on once Komodo reached the 6 piece Syzygy draw. Our experience with MTCS nn engines is they do not play the endgame as well as traditional programs. We also discussed Syzygy with the lc0 operator and he said they had not implement them yet. If they had them, we would have proposed a draw ourselves long before since why play on when we know the result? When you know little about a new opponent, the safest thing to do is play on."

Note Lefler claiming: "if Leela (knew it was draw) we would have proposed it long before ..."

Yanquis1972, Talkchess: "you say the board position implied no chance of a win, but that isn't how leela interpreted it. it makes sense to see if leela can draw against perfect play. this is the same thing an impartial operator would do out of sheer curiosity."

I was watching the game online and had a 1080 version of Leela running analysis in the background. I was posting into Lc0 forum and wrote this (actually just at the moment Leela lost on time): 

"Komodo will have EGTBs presumably. Leela has no EGTB but the neural network eval in this P on R-file endgame will act as a perfectly good EGTB. So, consider it a draw, both sides have enough chess "knowledge" at this stage of the game"

It was obvious for some many moves that Leela (her neural net eval) knew that taking the black a4 pawn, giving herself a passed a-file pawn was drawn. At the same time Leela thought if she didn't take the pawn the KRPvKBP, her neural net eval was about +3 pawns. Obviously all that was going to happen in this game was that Leela was going to play to keep the black king cutoff, but she was never going to take the a-pawn. Therefore draw after a lot of useless shuffling around.

And that is what happened. Leela showed +3 and her main lines NEVER showed a capture on a4.

So, I'm a competent (ex)chess programmer, I could tell immediately in three way a) the endless shuffling not capturing on a4, b) the evaluations, c) the lack of the capture move in the main lines that Leela had, in effect a neural net EGTB for KPK where P is on the a-file.

If I could tell so quickly, then Lefler, the programmer, wondering around, talking to the Leela operator, talking to his own operator, and obviously interested in the result, can hardly failed to have noticed it too.

Lefler's reason for playing on (basically that anything could happen because Leela has no EGTB, otherwise he'ld have taken the draw way earlier) is disingenuous to say the least.

[Tryfon Gavriel]

Hi all,

We "bombed" for sure in the World championship but we can learn from it. In particular, I wonder if we should try and establish some specific "tournament checklists" - so maybe for this particular tournament next time, we should consider using tablebases - because it does seem to me a massive burden on the operator to have to put in the moves especially for these 100+ move games. We could have checklists established in advance for all different weird and wonderful tournaments and specific rules of specific tournaments. 

I mention "checklists" because traditionally they were very useful avoiding embarrassing human-machine blunders as described here:

See this article about checklists:

https://www.artofmanliness.com/articles/the-power-of-checklists/

"B-17 Bombers. You’re likely familiar with the iconic B-17 “Flying Fortress” Bomber. But did you know if it weren’t for a simple checklist, it never would have gained its renown in WWII? In the 1930s, the U.S. Army Air Corps held a competition for airplane manufacturers vying to secure a contract to build the military’s next long-range bomber. Boeing produced a plane that could carry five times as many bombs as the army requested, and flew faster and further than previous bombers. On the day Boeing demonstrated its Flying Fortress, the plane lifted off the tarmac, stalled at 300 feet, and then crashed in a fiery explosion.

Investigators discovered the crash wasn’t caused by a mechanical malfunction, but rather pilot error. The problem was while the new bomber could carry more and fly faster and further than any other bomber in history, it was also an extremely complex plane to operate. To fly it, a pilot had to pay attention to four different engines, retractable landing gear, wing flaps, electric trim tabs, and much, much more. Because the pilot was so preoccupied with all these different systems, he forgot to release a new locking mechanism on the elevator and rudder controls. Overlooking something so simple killed the two men at the helm.

"

We should consider each tournament and its nuances and create a certain leela implementation and associated checklists for that implementation in my view.

This is the sort of silly stuff we can avoid.

The good news for me is that Leela held her own on the board - which is the really important thing. She wasn't positionally or tactically crushed etc. She held the position.

Cheers, K

[M]

From http://icga.org/

"Round 3 has started. Unfortunately Leela Chess Zero will not be playing at this moment, due to unforeseen problems. They are looking for another operator to finish the tournament. If anybody is available to operate, please contact both the Leela team and the ICGA organisation (webm...@icga.org)."

What is going on there?

[John D]

screw it. i'll do it. email me the plane tickets, departing o'hare. lodging for the first night would help since this is quite sudden.

[Martin Renneke]

@cyril has to went home, because of an accident in his family. I hope an other operator is on the way to Stockholm already. 

[Tryfon Gavriel]

If we can't get an operator out there - can we politely encourage on the Komodo team members there to be our operator - especially given the unfortunate 1st round game?

I would video annotate some of Komodo's wins if they could do us that favour. Don't know if they read the forums etc.

Is that a remote possibility? Or is Leela now going to default games. 

[Alexander Lyashuk]

We will have an operator for a second game today!

Also hardware will be different. Instead of 8x VT100, it will be 2x GTX1080ti (same computer that played TCEC).

--

You received this message because you are subscribed to the Google Groups "LCZero" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lczero+un...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/lczero/407d54fb-e153-46d3-b470-4f7e7f50d80f%40googlegroups.com.

[M]

That's bad news. Lets hope for the best for his family.

[Alexander Lyashuk]

fyi the game has started.