TalkChess.com

In short, version 0.37 uses Ordo model by Gaviota, while version 0.38 uses Bayeselo model from Rémi Coulom.

But Ordo is much more popular than Bayeselo, so Mr. Phhnguyen might consider switching from Bayeselo to Ordo.

The algorithm of Ordo can be found here [https://sites.google.com/site/gaviotachessengine/ordo] in the ''Manual'' menu

More popular yes, and maybe that is reason enough to switch , but it is not universally accepted as better, just different.

rongle99 wrote: ↑Fri Dec 25, 2020 6:38 am In short, version 0.37 uses Ordo model by Gaviota, while version 0.38 uses Bayeselo model from Rémi Coulom.

But Ordo is much more popular than Bayeselo, so Mr. Phhnguyen might consider switching from Bayeselo to Ordo.

The algorithm of Ordo can be found here [https://sites.google.com/site/gaviotachessengine/ordo] in the ''Manual'' menu

No, it was LOS but not Ordo’s Elo. You can see my implementation of LOS in Banksia - the open source one, it is so simple and quite far from Ordo’s one.

BTW, I have been considering to implement both Ordo and Bayseselo methods thus users can take anyone.

Modern Times wrote: ↑Fri Dec 25, 2020 7:17 am More popular yes, and maybe that is reason enough to switch , but it is not universally accepted as better, just different.

The main reason I have implemented Bayeselo first is that some famous chess websites use that method

Really appreciate you will implement both, thank you for your hard work

And I also like to see the reports in the PGN file, Time is now reported in milliseconds and not seconds, this is really very good, coz sometimes we use a very small tc (say 0.05s/move) If the report is still in seconds, then it will read as zero (0) in the PGN file.

Happy Christmas

Do Banksia GUI take castling rights into consideration in case of threefold repetition ?

Karpov versus Miles, 1986

[d]r3kb1r/5ppp/4p3/1N6/4P3/8/Pn1BK1PP/R6R b kq - 1 22

Karpov claims draw with his third move Nb5. But the draw by repetion rule also considers the castling availability. At the first Karpov's Nb5, Anthony Miles still had queen side castling movement. With the second Karpov's Nb5, Anthony Miles had no more the castling available. So, the first position is different of the second by this reason. By a mistake, Karpov claims the draw after his thirt Nb5 move, but the position was repeated only twice.

It need to be repeated four times.

well, Banksia did not automaticly update engines ELO

i expected that this will be update and replace the ELO ratings of the engines right after finished a tournament.

phhnguyen wrote: ↑Fri Dec 25, 2020 5:23 am

Rom77 wrote: ↑Fri Dec 25, 2020 3:19 am Why is there such a big difference? Is it possible to enter an option for calculating the classic Elo?

v0.37 :


v0.38 :

Previous versions of 0.38 were not Elo but LOS (Likelihood of Superiority).

https://www.chessprogramming.org/Match_ ... uperiority

When discussing we mostly talk about Elo but not LOS even LOS can be used to compare engines.

LOS is much easier to implement (that is why BSG has it so soon) when Elo/BaysesElo required me a lot of effort to understand and implement.

But 93.4 is not LOS. The number 93.4 is Elo. Here is the calculation on a regular Elo calculator:

Could you please give the option to set the background of the Score Graph?

I want to set the lines in the Score Graph like this [https://i.postimg.cc/g2mMYXR0/Screenshot-2.png], but the White Line in the Score Graph is not visible, coz the background is white.

I am using Banksia GUI lately and it is very good for doing tournaments with different time control for each player.

The concurrency is another excellent feature of Banksia if you want to play several games simultaneously using a few cores per engine.

The problem of the concurrency is that you can't use it with a GPU engine. In order to do it possible I propose to implement a feature in Banksia
defining in each engine a variable that indicates whether the engine is GPU type or it is CPU type.
In the Overall options the "Cores" option must be splitted in "Cores per CPU engine" and "Cores per GPU engine" (typically equal to 2).
The games of the tournament must be organized in a way that only one GPU engine must use the GPU and when the game is finished, the GPU can be used by another GPU engine. So the concurrency of the GPU engines is 1 while the user can set the concurrency of the CPU engines greater than one.

One way to organize a tournament is to play the GPU engine games first, using the remaining cores with the concurrency set for the CPU engines. At the end of the GPU games the computer can use all the CPU engines cores plus the cores reserved for the GPU engines previously to finish the games with CPU engines.

I know that if we want to play a tournament with many GPU engines, the concurrency would lose efficiency because the GPU would always be busy while the CPU would have unused cores.

With this new feature interesting experiments can be done!

I want to thank Nguyen Hong Pham for creating Banksia GUI and wish him the best for him and his family.