Re: Empirically Logistic ELO model better suited than Gaussi
Posted: Thu Jul 14, 2016 8:25 pm
Come on, Kai, preserve your hand for better things....
Hosted by Your Move Chess & Games - chessusa.com
Those Etruscans with their Lars Porsenna are nasty things. I have to kill them.fern wrote:Come on, Kai, preserve your hand for better things....
There are some interesting theories about the Hittites, whose civilization and kingdom vanished mysteriously from history. Some say they actually migrated, possibly to Europe. The same is said about the Assyrians.fern wrote:Yes, once and again it is discovered that old legends and traditions has more than one core or two of truth.
AS You know Anatolia was the cradle of some very old civilizations, Hittites between them, lot of people movements, etc.
And remember the old histories about the "sea peoples" which were agents of destruction around 12 century before Christ. Egyptians talked a lot about them.
We need urgently to put all this clear, a time machine is necessary. You look so technically gifted that perhaps you could make one.
Pompei -or at least his severed head- send his regards
The difficulty with the draw model is that it's not really a solid model, it depends on engine and time control. Nevertheless, for selected engines, at some short time controls, I managed to rule out the Rao-Kupper model as used in BayesELO at p<0.001 level. Left are Davidson and Glenn-David models. Glenn-David is derived from Gaussian ELO model, and as previously shown, engines seem to obey Logistic ELO model, so Davidson draw model based on Logistic would be a safer bet. I post here the results for three engines, Texel, Andscacs and Stockfish. Observe that each of these engines has a bit differently normalized draw model, showing the engine dependency of the model. But all results seem to suggest that Rao-Kupper is ruled out. I used Draw/Win ratio as function of Win to fit with models. I think more comprehensive statistical tests of good fit from say CCRL database are difficult to perform, because the data is too noisy, albeit the database is large.