TCEC Division 3 results simulator

[chrisw]

I think the leela fans prefer "leela was disconnected", which if it is the case may allow the game to be replayed according to the rules.

I want to add that also Hannibal 'disconnected' just a few games before this game.
No one questions though, if it really was its fault. (missing fan hype?)
May be Edsel can say something on this matter after he had seen the logs.

This is typical for organisations with interpretable rules. It's very unlikely there's any bias against Hannibal, but nobody made any noises so they went with the obvious "disconnect" decision.

There's a lot of noise about LC0 disconnect, so it gets checked and double checked and tripled checked and so on. I guess we can assume there won't be any bias when they decide it was the server's fault.

Question is, will they make equivalent effort to now reexamine the Hannibal game.

[Laskos]

looks like they will call it a server disconnect, so - replay game presumably.

Do they play it now, or at the end?

I don't know. It seems, if it gets replayed, the main opponent is Arasan, by points and tiebreak advantage, as of now, equivalent to 2 full points. So, no replay - 8% to Lc0, replay - 20% to Lc0 (bad tiebreak against Arasan). Pedone appears marginally, only if Lc0 plays mediocrely, in which case Lc0 anyway won't qualify. All three have a game against Ethereal beast, and only Arasan against Bobcat mice.

[Milos]

I think the leela fans prefer "leela was disconnected", which if it is the case may allow the game to be replayed according to the rules.

I want to add that also Hannibal 'disconnected' just a few games before this game.
No one questions though, if it really was its fault. (missing fan hype?)
May be Edsel can say something on this matter after he had seen the logs.

This is typical for organisations with interpretable rules. It's very unlikely there's any bias against Hannibal, but nobody made any noises so they went with the obvious "disconnect" decision.

There's a lot of noise about LC0 disconnect, so it gets checked and double checked and tripled checked and so on. I guess we can assume there won't be any bias when they decide it was the server's fault.

Question is, will they make equivalent effort to now reexamine the Hannibal game.

Well whole tournament is just oriented towards money generation and nothing stops organizer in trying to get as much ppl to watch as possible. It includes breaking all of tournament rules, selective application of them and apparent injustices. One reason is to create controversy, but another more important one is not to lose huge crowed of NN engine fanboys, Lc0 in the first place and if possible DX too. In Div4 Ivanohoe was playing on a single core as it was an obviously bad compile provided by Kranium and after a lot of protest they just decided to ignore it. Allowing Ivanhoe on 43 cores would mean one NN would have to stay in Division 4 and almost certainly no chance for any NN engine to promote further from Div 3.
Now they seems to allowed a change in configuration parameters of Lc0 after more than half of the games were played not because Lc0 was crashing but performing bad. This has been done secretly to avoid humiliation of being criticised for such a blunt breaking of rules. To anyone that actually followed games or compared the logs the apparent change in time management is quite obvious (just plotting time usage figure is enough).
Finally, they seems to have ruled a game where Lc0 hanged (and it happened in the past too, so it is not the first time Lc0 hangs) as a "server disconnect" again in desperate attempt to somehow push Lc0 to Div3 and not lose so many fanboys that generate traffic on the website and accordingly bring money to organiser.
But unfortunately for them, even with this replay they don't stand much chance. Your simulator is quite correct in that respect, much better than some gut feelings obviously. So even with a repeated game Lc0 stands less than 10% of chance to go further which in some sense is kind of poetic justice in the end.

[frankp]

"Yes the game will be replayed. The logs made the reason very clear: 47519116 <lc0 16.10520(22): FATAL ERROR: Network error: Software caused connection abort"

Taken from the leela forum, in answer to the question will the hannibal v leela game be replayed.

[Uri Blass]

I think the leela fans prefer "leela was disconnected", which if it is the case may allow the game to be replayed according to the rules.

I want to add that also Hannibal 'disconnected' just a few games before this game.
No one questions though, if it really was its fault. (missing fan hype?)
May be Edsel can say something on this matter after he had seen the logs.

This is typical for organisations with interpretable rules. It's very unlikely there's any bias against Hannibal, but nobody made any noises so they went with the obvious "disconnect" decision.

There's a lot of noise about LC0 disconnect, so it gets checked and double checked and tripled checked and so on. I guess we can assume there won't be any bias when they decide it was the server's fault.

Question is, will they make equivalent effort to now reexamine the Hannibal game.

Well whole tournament is just oriented towards money generation and nothing stops organizer in trying to get as much ppl to watch as possible. It includes breaking all of tournament rules, selective application of them and apparent injustices. One reason is to create controversy, but another more important one is not to lose huge crowed of NN engine fanboys, Lc0 in the first place and if possible DX too. In Div4 Ivanohoe was playing on a single core as it was an obviously bad compile provided by Kranium and after a lot of protest they just decided to ignore it. Allowing Ivanhoe on 43 cores would mean one NN would have to stay in Division 4 and almost certainly no chance for any NN engine to promote further from Div 3.
Now they seems to allowed a change in configuration parameters of Lc0 after more than half of the games were played not because Lc0 was crashing but performing bad. This has been done secretly to avoid humiliation of being criticised for such a blunt breaking of rules. To anyone that actually followed games or compared the logs the apparent change in time management is quite obvious (just plotting time usage figure is enough).
Finally, they seems to have ruled a game where Lc0 hanged (and it happened in the past too, so it is not the first time Lc0 hangs) as a "server disconnect" again in desperate attempt to somehow push Lc0 to Div3 and not lose so many fanboys that generate traffic on the website and accordingly bring money to organiser.
But unfortunately for them, even with this replay they don't stand much chance. Your simulator is quite correct in that respect, much better than some gut feelings obviously. So even with a repeated game Lc0 stands less than 10% of chance to go further which in some sense is kind of poetic justice in the end.

Not so sure that lc0 has small chances to promote and the question is what is the rating improvement of lc0 in the middle of the tournament

Note that if lc0 does not promote even after replaying lc0's game when lc0 crashed they can try repeat also the games when nemorino crashed so maybe nemorino can beat arasan and if it does not help they can repeat also the first games of lc0.
If they want lc0 to promote they can also decide to repeat arasan-nemorino when
They can also repeat earlier games of lc0 when lc0 was performing bad.

[chrisw]

I think the leela fans prefer "leela was disconnected", which if it is the case may allow the game to be replayed according to the rules.

I want to add that also Hannibal 'disconnected' just a few games before this game.
No one questions though, if it really was its fault. (missing fan hype?)
May be Edsel can say something on this matter after he had seen the logs.

This is typical for organisations with interpretable rules. It's very unlikely there's any bias against Hannibal, but nobody made any noises so they went with the obvious "disconnect" decision.

There's a lot of noise about LC0 disconnect, so it gets checked and double checked and tripled checked and so on. I guess we can assume there won't be any bias when they decide it was the server's fault.

Question is, will they make equivalent effort to now reexamine the Hannibal game.

Well whole tournament is just oriented towards money generation and nothing stops organizer in trying to get as much ppl to watch as possible. It includes breaking all of tournament rules, selective application of them and apparent injustices. One reason is to create controversy, but another more important one is not to lose huge crowed of NN engine fanboys, Lc0 in the first place and if possible DX too. In Div4 Ivanohoe was playing on a single core as it was an obviously bad compile provided by Kranium and after a lot of protest they just decided to ignore it. Allowing Ivanhoe on 43 cores would mean one NN would have to stay in Division 4 and almost certainly no chance for any NN engine to promote further from Div 3.
Now they seems to allowed a change in configuration parameters of Lc0 after more than half of the games were played not because Lc0 was crashing but performing bad. This has been done secretly to avoid humiliation of being criticised for such a blunt breaking of rules. To anyone that actually followed games or compared the logs the apparent change in time management is quite obvious (just plotting time usage figure is enough).
Finally, they seems to have ruled a game where Lc0 hanged (and it happened in the past too, so it is not the first time Lc0 hangs) as a "server disconnect" again in desperate attempt to somehow push Lc0 to Div3 and not lose so many fanboys that generate traffic on the website and accordingly bring money to organiser.
But unfortunately for them, even with this replay they don't stand much chance. Your simulator is quite correct in that respect, much better than some gut feelings obviously. So even with a repeated game Lc0 stands less than 10% of chance to go further which in some sense is kind of poetic justice in the end.

I also intuit that "something has changed" part way through the event.

[chrisw]

If LC0 draws it's current game with Ethereal (quite likely), and Arasan loses to Ethereal (quite possible) and LC0 gets the disconnect replay and wins, then LC0 is in joint second place.

[Guenther]

I also intuit that "something has changed" part way through the event.

same settings as before:

Code: Select all

Protocol=uci; Move time overhead in milliseconds=2000; Time weight curve peak ply=37; Aggressive smart pruning threshold=0.604; Scale thinking time=3.05;

[Nay Lin Tun]

I think the leela fans prefer "leela was disconnected", which if it is the case may allow the game to be replayed according to the rules.

I want to add that also Hannibal 'disconnected' just a few games before this game.
No one questions though, if it really was its fault. (missing fan hype?)
May be Edsel can say something on this matter after he had seen the logs.

This is typical for organisations with interpretable rules. It's very unlikely there's any bias against Hannibal, but nobody made any noises so they went with the obvious "disconnect" decision.

There's a lot of noise about LC0 disconnect, so it gets checked and double checked and tripled checked and so on. I guess we can assume there won't be any bias when they decide it was the server's fault.

Question is, will they make equivalent effort to now reexamine the Hannibal game.

Well whole tournament is just oriented towards money generation and nothing stops organizer in trying to get as much ppl to watch as possible. It includes breaking all of tournament rules, selective application of them and apparent injustices. One reason is to create controversy, but another more important one is not to lose huge crowed of NN engine fanboys, Lc0 in the first place and if possible DX too. In Div4 Ivanohoe was playing on a single core as it was an obviously bad compile provided by Kranium and after a lot of protest they just decided to ignore it. Allowing Ivanhoe on 43 cores would mean one NN would have to stay in Division 4 and almost certainly no chance for any NN engine to promote further from Div 3.
Now they seems to allowed a change in configuration parameters of Lc0 after more than half of the games were played not because Lc0 was crashing but performing bad. This has been done secretly to avoid humiliation of being criticised for such a blunt breaking of rules. To anyone that actually followed games or compared the logs the apparent change in time management is quite obvious (just plotting time usage figure is enough).
Finally, they seems to have ruled a game where Lc0 hanged (and it happened in the past too, so it is not the first time Lc0 hangs) as a "server disconnect" again in desperate attempt to somehow push Lc0 to Div3 and not lose so many fanboys that generate traffic on the website and accordingly bring money to organiser.
But unfortunately for them, even with this replay they don't stand much chance. Your simulator is quite correct in that respect, much better than some gut feelings obviously. So even with a repeated game Lc0 stands less than 10% of chance to go further which in some sense is kind of poetic justice in the end.

Not so sure that lc0 has small chances to promote and the question is what is the rating improvement of lc0 in the middle of the tournament

Note that if lc0 does not promote even after replaying lc0's game when lc0 crashed they can try repeat also the games when nemorino crashed so maybe nemorino can beat arasan and if it does not help they can repeat also the first games of lc0.
If they want lc0 to promote they can also decide to repeat arasan-nemorino when
They can also repeat earlier games of lc0 when lc0 was performing bad.

Lc0 crash was investigated as network failure( not engine fault). Nemorino crash was due to engine error (failure ). So the games wont be replayed. They are different scanerios.

[frankp]

Apparently, one big change is the temperature in the Netherlands decreased, so the gpu are not overheating - as much.

"GPU behaving better, though GPU1 is still hitting thermal limits at times. I may need to reduce power to lower than 90%"
reported on forum just now