TCEC Division 3 results simulator

[chrisw]

[d]2r5/1p2Q3/p1rp2p1/P1P1n1k1/4P3/1P5P/1B5P/6K1 b - - 2 40

Arasan could actually screw up here and only draw, unlikely, but possible

[chrisw]

Division 3 decided. Ethereal and Arasan go through. Well done Ethereal programmer, that's a really nice speculative evaluator you programmed there, some of the main line scores were seriously impressive, almost neural network-like. And well done John Dart, well deserved!

Meanwhile, sim promotion predictions for Division 2 (based on some possibly dodgy elo stats)
Ethereal 89%
Gull 64%
Nirvana 35%
Xiphos 3%
Arasan 4%
Vajolet 2%
Texel 1%
ChainBrain 0%

Code: Select all

Division 3,
Engine,   Tournament Elo, Initial Elo, First, First two
Ethereal       3314         3300       1.000  1.000
Arasan         3146         3124       0.000  1.000
Pedone         3117         3092       0.000  0.000
DeusX          3129         3130       0.000  0.000
Nemorino       3093         3099       0.000  0.000
Hannibal       3120         3158       0.000  0.000
Bobcat         2978         3029       0.000  0.000
lc0            3132         3100       0.000  0.000

Division 2,
Engine,   Tournament Elo, Initial Elo, First, First two
Ethereal       3314         3314       0.650  0.890
Gull           3263         3263       0.236  0.644
Nirvana        3229         3229       0.100  0.359
Xiphos         3138         3138       0.005  0.033
Arasan         3146         3146       0.005  0.040
Vajolet        3128         3128       0.003  0.022
Texel          3111         3111       0.001  0.012
ChessBrainVB   2983         2983       0.000  0.000

[Milos]

Division 3 decided. Ethereal and Arasan go through. Well done Ethereal programmer, that's a really nice speculative evaluator you programmed there, some of the main line scores were seriously impressive, almost neural network-like. And well done John Dart, well deserved!

Are you sure it is decided?
They can still decide to replay all the NN games before the thermal "issue". Or they can change tiebreak rules. Or they could give a wild card to Lc0.
Judging how they respect the rules and how tournament is run all those scenarios certainly have non-zero chance of happening .

[Guenther]

Division 3 decided. Ethereal and Arasan go through. Well done Ethereal programmer, that's a really nice speculative evaluator you programmed there, some of the main line scores were seriously impressive, almost neural network-like. And well done John Dart, well deserved!

Meanwhile, sim promotion predictions for Division 2 (based on some possibly dodgy elo stats)
Ethereal 89%
Gull 64%
Nirvana 35%
Xiphos 3%
Arasan 4%
Vajolet 2%
Texel 1%
ChainBrain 0%

Code: Select all

Division 2,
Engine,   Tournament Elo, Initial Elo, First, First two
Ethereal       3314         3314       0.650  0.890
Gull           3263         3263       0.236  0.644
Nirvana        3229         3229       0.100  0.359
Xiphos         3138         3138       0.005  0.033
Arasan         3146         3146       0.005  0.040
Vajolet        3128         3128       0.003  0.022
Texel          3111         3111       0.001  0.012
ChessBrainVB   2983         2983       0.000  0.000

FWIW here some data from CCRL plus notes and rough additional rating improvement expectations
[my guess given as ++/+/(+)]for newer versions.

Code: Select all

#Name          CCRL40/40 4CPU   Version  Improv.(Guess)
Ethereal       3295             10.55    ++ 
Nirvana        3187              2.4     
Gull           3184              3.0    
Xiphos         3181              0.3     ++
Texel          3145              1.07    +
ChessbrainVB   3128              3.68    (+)
ArasanX        3124             20.41    ++
Vajolet2       3101              2.5     +

*notes:
- Nirvana and Gull not under development anymore
- inofficial Gull 3.1 is below official 3.0 for given rating source

[Uri Blass]

Division 3 decided. Ethereal and Arasan go through. Well done Ethereal programmer, that's a really nice speculative evaluator you programmed there, some of the main line scores were seriously impressive, almost neural network-like. And well done John Dart, well deserved!

Are you sure it is decided?
They can still decide to replay all the NN games before the thermal "issue". Or they can change tiebreak rules. Or they could give a wild card to Lc0.
Judging how they respect the rules and how tournament is run all those scenarios certainly have non-zero chance of happening .

Note that the majority voted for replaying all the NN games(I voted against it).

https://www.strawpoll.me/16282056/r

[chrisw]

Division 3 decided. Ethereal and Arasan go through. Well done Ethereal programmer, that's a really nice speculative evaluator you programmed there, some of the main line scores were seriously impressive, almost neural network-like. And well done John Dart, well deserved!

Meanwhile, sim promotion predictions for Division 2 (based on some possibly dodgy elo stats)
Ethereal 89%
Gull 64%
Nirvana 35%
Xiphos 3%
Arasan 4%
Vajolet 2%
Texel 1%
ChainBrain 0%

Code: Select all

Division 2,
Engine,   Tournament Elo, Initial Elo, First, First two
Ethereal       3314         3314       0.650  0.890
Gull           3263         3263       0.236  0.644
Nirvana        3229         3229       0.100  0.359
Xiphos         3138         3138       0.005  0.033
Arasan         3146         3146       0.005  0.040
Vajolet        3128         3128       0.003  0.022
Texel          3111         3111       0.001  0.012
ChessBrainVB   2983         2983       0.000  0.000

FWIW here some data from CCRL plus notes and rough additional rating improvement expectations
[my guess given as ++/+/(+)]for newer versions.

Code: Select all

#Name          CCRL40/40 4CPU   Version  Improv.(Guess)
Ethereal       3295             10.55    ++ 
Nirvana        3187              2.4     
Gull           3184              3.0    
Xiphos         3181              0.3     ++
Texel          3145              1.07    +
ChessbrainVB   3128              3.68    (+)
ArasanX        3124             20.41    ++
Vajolet2       3101              2.5     +

*notes:
- Nirvana and Gull not under development anymore
- inofficial Gull 3.1 is below official 3.0 for given rating source

Thanks for those, I was just actually in the process of making some adjusts already:

"Gull", 0, 3184,
"Texel", 0, 3145,
"Xiphos", 0, 3181,
"ChessBrainVB", 0, 3128,
"Nirvana", 0, 3187,
"Vajolet", 0, 3101,
"Ethereal", 0, 3335,
"Arasan", 0, 3174,

The same as yours, but I gave Ethereal some pts to adjust for its tournament performance in Div3, likewise Arasan. 40 to Ethereal and 50 to Arasan.
Should I reduce Gull a little? And Nirvana?

[chrisw]

Division 3 decided. Ethereal and Arasan go through. Well done Ethereal programmer, that's a really nice speculative evaluator you programmed there, some of the main line scores were seriously impressive, almost neural network-like. And well done John Dart, well deserved!

Are you sure it is decided?
They can still decide to replay all the NN games before the thermal "issue". Or they can change tiebreak rules. Or they could give a wild card to Lc0.
Judging how they respect the rules and how tournament is run all those scenarios certainly have non-zero chance of happening .

Note that the majority voted for replaying all the NN games(I voted against it).

https://www.strawpoll.me/16282056/r

Well, you're sensible, whilst the poll was promoted at Leela Discord as: "Straw poll to convince Anton, go vote", so that's a majority that can be safely ignored.

[Guenther]

Thanks for those, I was just actually in the process of making some adjusts already:

"Gull", 0, 3184,
"Texel", 0, 3145,
"Xiphos", 0, 3181,
"ChessBrainVB", 0, 3128,
"Nirvana", 0, 3187,
"Vajolet", 0, 3101,
"Ethereal", 0, 3335,
"Arasan", 0, 3174,

The same as yours, but I gave Ethereal some pts to adjust for its tournament performance in Div3, likewise Arasan. 40 to Ethereal and 50 to Arasan.
Should I reduce Gull a little? And Nirvana?

Note that the adjustment is also there in my data by the given column 'Improv. Guess', just that you have to add some rating
according to the ++/+/(+) values, which means add quite some for Ethereal, Arasan and Xiphos and some for Texel and Vajolet2,
may be a tiny bit for ChessbrainVB and nothing to Nirvana and Gull.

[chrisw]

Thanks for those, I was just actually in the process of making some adjusts already:

"Gull", 0, 3184,
"Texel", 0, 3145,
"Xiphos", 0, 3181,
"ChessBrainVB", 0, 3128,
"Nirvana", 0, 3187,
"Vajolet", 0, 3101,
"Ethereal", 0, 3335,
"Arasan", 0, 3174,

The same as yours, but I gave Ethereal some pts to adjust for its tournament performance in Div3, likewise Arasan. 40 to Ethereal and 50 to Arasan.
Should I reduce Gull a little? And Nirvana?

Note that the adjustment is also there in my data by the given column 'Improv. Guess', just that you have to add some rating
according to the ++/+/(+) values, which means add quite some for Ethereal, Arasan and Xiphos and some for Texel and Vajolet2,
may be a tiny bit for ChessbrainVB and nothing to Nirvana and Gull.

Okay, got it now. I gave Ethereal and Arasan +40 and +50, so will give Xiphos +45, Texel and Vajolet +20, and ChessBrain 5. Does that figure?

"Gull", 0, 3184,
"Texel", 0, 3145+20,
"Xiphos", 0, 3181+45,
"ChessBrainVB", 0, 3128+5,
"Nirvana", 0, 3187,
"Vajolet", 0, 3101+20,
"Ethereal", 0, 3335,
"Arasan", 0, 3174,

The first runs of the Div2 Sim will be without any game data, so will be entirely based on elo rank. Those figures would predict Ethereal and Xiphos through.

[Guenther]

Okay, got it now. I gave Ethereal and Arasan +40 and +50, so will give Xiphos +45, Texel and Vajolet +20, and ChessBrain 5. Does that figure?

"Gull", 0, 3184,
"Texel", 0, 3145+20,
"Xiphos", 0, 3181+45,
"ChessBrainVB", 0, 3128+5,
"Nirvana", 0, 3187,
"Vajolet", 0, 3101+20,
"Ethereal", 0, 3335,
"Arasan", 0, 3174,

Sounds good enough for an educated guess