AUGUST 2025 BLITZ TOURNAMENT

[Joost Buijs]

About 95 to 110

Hi Sidiki,

Thank you, that's interesting.

Joost

[Sidiki]

Hi Joost,

No problem, the tournament is running. I uploaded the 2 first games round 7.

Friendly, Sidiki

[Sidiki]

Hi everybody,

Round 8 finished.
https://results.fmjd.org/tournaments/2025/f_1228/

Friendly, Sidiki.

[Sidiki]

Hi everybody,

Round 10 finished.
https://results.fmjd.org/games/#/round/2025[@SL]f_1228/10/22

Thanks.

Friendly, Sidiki

[Sidiki]

Hi everyone,

Sorry for the matches today.
I'm in the middle of moving, probably tomorrow we'll have at least one round, thank you all.

Friendly, Sidiki

[Sidiki]

Hi everybody,

Round 12 finished.
https://results.fmjd.org/games/#/round/2025[@SL]f_1228/12/22

Friendly, Sidiki.

[BertTuyt]

Sidiki, thanks for organizing.
Several (interesting) observations.

It seems that these days Ares and Kingsrow are the 2 strongest programs.
Now both leading (after 12 rounds) with 2 points more then the followers.
Clearly visible that MobyDam, GWD and Tornado become stronger and stronger, as their developers are still actively improving their programs.
If I counted correctly, 3 programs still use a HCE (Hand Crafted Evaluation).
These are: Horizon, Flits and Truus.
The other 9 use an evaluation based upon some form of ML (Machine Learning).
So far in the tournament (but things can change), the top 9 always draws against each other, and the wins are always against the 3 HCE programs.

Based on this can think about the future of Computer Draughts.
First of all it is my belief (but I can be wrong), that with tournament conditions, large endgame databases, and muti-core search (16-32 cores), there is little ELO gain possible beyond the current Kingsrow/Ares performance. My estimation maximum 10 ELO, but again I hope some-one will prove I'm wrong.

For better differentiation we needs more games, as with a limited number and small differences in strength there is always some statistics involved.
One can also stop improving the strength of a program and focus on winning against the weaker players. Not sure how to program this, but I'm sure there are ways to do so (think Sjende Blyn has implemented this, but Jelle could tell us more). But this strategy fails if we would have 12 programs in the tournament, all with ML based evaluation functions, and multi-threaded search.

Also an option to play games with very small time controls and limited hardware settings. As a basic test, I always do a 158 DXP match with 1 minute/game, a 6p DB, and 1 core for the search.
Short time controls (and maybe without pondering) could also enable more games, but then one should really "automate" a tournament.

And last but not least we could change the rules, go to Killer Draughts or something else...

But for me stopping with Computer Draughts is not (yet) an option

Bert

[Joost Buijs]

The (excessively) high number of draws is indeed a major problem in International Draughts.

First of all, it makes Machine Learning (ML) above a certain level virtually impossible, because a program that only plays draws doesn't learn anything new.

Secondly, it makes tournaments less interesting, because it's clear in advance that games between strong programs will always end in a draw.

The only way to break this pattern is to change the rules of the game in such a way that the likelihood of a draw decreases.

‘Killer Draughts’ is one possibility. Adapting the move generator for it is easy, but generating a new position database from scratch, and recalculating the neural network plus the endgame databases takes an enormous amount of time — and you really have to be motivated to go through with that.

[Sidiki]

Hi Joost,

that's correct.
Round 13 finished.

https://results.fmjd.org/games/#/round/2025[@SL]f_1228/13/22

Friendly, Sidiki

[Sidiki]

Hi everybody,

Round 16 finished

https://results.fmjd.org/games/#/round/2025[@SL]f_1228/16/22

Friendly, Sidiki

[Sidiki]

Hi everybody,

Round 18 finished.

https://results.fmjd.org/games/#/round/2025[@SL]f_1228/17/22

Friendly, Sidiki