Unofficial Olympics Among Draughts Programs

[Krzychumag]

On days 13 and 14 november 2012 year in Poland ware played the unofficial Olympics into international draughts. It in tournament started seven programmes. First place occupied programme Dragon, second place program Kingsrow but the third place program Truus .The bosom congratulations for all participants in the Olympics. I am inviting cordially the photograph and information about the Olympics downloadable.


Unofficial Olympics Among Draughts Programs - Word

Unofficial Olympics Among Draughts Programs - format pdn

Unofficial Olympics Among Draughts Programs - Foto



From respect
Grzelak Krzysztof

[Catherine]

Congratulations to Dragon V4.12 and Michel of course. I thung that is the first computer tournament that Kryztof made that stimulate Michel to improve the power of Dragon. With this result( without the 158 dxp games] is it possible to said that the actual Dragon is the best program in the world?
I tested it against Truus ; Flits; Damage 2011 and Dam 2.2 and i see that it is most stronger than ever.
Catherine.

[Krzychumag]

As regards the Olympics, am dividing the opinion Bert'a - it is necessary to have much with teasel in such championships. And if only to say the program is best it had to in order me to do championships with other programs of the type - Damage, Damy and other programmes. But how I know neither Bert Tuyt neither Gérard TAILLE they aren't making their programs available in the unabridged version.

[MichelG]

Thanks for running the tournament

Congratulations to Dragon V4.12 and Michel of course. I thung that is the first computer tournament that Kryztof made that stimulate Michel to improve the power of Dragon. With this result( without the 158 dxp games] is it possible to said that the actual Dragon is the best program in the world?
I tested it against Truus ; Flits; Damage 2011 and Dam 2.2 and i see that it is most stronger than ever.
Catherine.

I am glad dragon does this well on these tournaments, but because of the relatively small number of games, you can't conclude dragon is the best.

Me and Ed played a dxp match against kingsrow, and kingsrow has the upper hand.

Nevertheless, i am still making steady progress in improving dragon.

[Catherine]

Michel, is it possible to have the games against Kingsrow and against Damage.
In a topic you said that you worked on the speed of your alpha beta procedure. My question is: Is it possible to build an algorithm which can generate the perfect game?
I suppoze that Kingsrow,if i consider what you said about dxp games, is the best because his algorithm generate a peRfect move than the other program or this is due to is high speed search alpha beta.

[BertTuyt]

Michel , could you share the results of your match against Kingsrow?
I post regular Damage-Kingsrow Match results , and step by step i try to reduce the gap....

Bert

[BertTuyt]

To my opinion the strength of Kingsrow is based on a very stable, and aggressive search routine with a stable solid evaluation.
Ed did a tremendous job, based on playing and observing many many (initially non-automatic) games.

Bert

[BertTuyt]

As i play multiple matches these days (see this forum), I often observe a sequence of more than 30 draws against Kingsrow during a 158 games match.
And more and more the number off non-draw games is smaller as 10.

I'm sure if other programmers start with implementing well documented concepts like MCP and LMR, then they will see similar behavior.
Add to this parallel search (just browse the Stockfish source code to get sufficient clues).

So my pessimistic view, in theory soon all programs could grow into the KingsRow type of strength.
From that point onwards, a normal tournament just becomes a lottery...
We really need to think how we get out of the Draw trap.

Maybe Ed can comment, but i believe that for checkers computer tournaments these days are non existing, it takes a few hundred games to see once decisive non-draw results.
And if all would implement the known search concepts, generate 7p - 8p DB's and use the power of all cores, than we face the same situation in International 10x10 Draughts.

Bert

[Krzychumag]

Bert have such request to You. Whether you can make available one's newest versions of program Damage. I am thinking whether Ed played some match with the pace of 20 minutes at any time for 75 moves. Whether who stab the age is playing such matches. I am thinking whether there is some meaning of playing such Olympics in the future.

[Krzychumag]

Bert as regards the code, I propose to use the commercial program Houdini 3.0.

[Catherine]

Bert, thaks for replies. Now I know any program can be improved using some technical tricks. Damage is absent to all the recents computers tournament. Why? And (not to do a comparison), but have you done a dxp games against Dragon and Damy? To finish, in your point of view, what are the strenghts of the actual Damage? Thank.
Catherine.

[MichelG]

Michel , could you share the results of your match against Kingsrow?
I post regular Damage-Kingsrow Match results , and step by step i try to reduce the gap....

Bert

Kingsrow took 62.1% of the score against dragon 4.1 (85 elo points, 259/696/22)

As i play multiple matches these days (see this forum), I often observe a sequence of more than 30 draws against Kingsrow during a 158 games match.
And more and more the number off non-draw games is smaller as 10.

I'm sure if other programmers start with implementing well documented concepts like MCP and LMR, then they will see similar behavior.
Add to this parallel search (just browse the Stockfish source code to get sufficient clues).

So my pessimistic view, in theory soon all programs could grow into the KingsRow type of strength.
From that point onwards, a normal tournament just becomes a lottery...
We really need to think how we get out of the Draw trap.

Bert

The open question is whether the current top programs are nearing perfection, meaning that they on avarage make less than 1 errornous move every 1000 or so turns (or 1 in 20 games)

It is possible to estimate how many errors the programs make; let one player play at 10 minutes/game, the other at 24 hours/game. If you still get draws, that means that the 10 minute player is playing fault-less and can't be improved upon.

If the 24 hours/game player wins often, that means that there is still room for improvement (you would just have to improve nodes/second by a factor 144)

One (temporary) solution would be to have the computers play at faster time controls, e.g 30 seconds per game.

[BertTuyt]

Bert, thaks for replies. Now I know any program can be improved using some technical tricks. Damage is absent to all the recents computers tournament. Why? And (not to do a comparison), but have you done a dxp games against Dragon and Damy? To finish, in your point of view, what are the strenghts of the actual Damage? Thank.
Catherine.

Good questions. I only play a tournament with Damage if i reached a specific development goal with my program. For example the last tournament I played , my program search was based on a parallel (YBWC) algorithm, and it was the first time I used the 7P DB.
My new goal is to beat Kingsrow in a 158 games match. As for me it doesn't make sense to become champion if im not able to beat the best program in the world, which is still KingsRow.
So if i reach that goal I will most likely participate in a next tournament.

As Damy is not available, i dont have a dxp match caparison.
I plan to generate the 6P DB for Dragon and then will play a match, most likely timing around XMas holiday.
As Damage now comes quite close to KingsRow although Damage plays with a 7P DB and the KingsRow version I use has "only" 6P.
I post the match results quite regular, and I'm now around 50% (or slightly less) for Damage.
Based on the Dragon - Kingsrow match results I guess Damage would win a match against Dragon, but the proof of the pudding is always in the eating.

Bert

[Krzychumag]

Bert and why you don't want to make the latest version available Damage.

[BertTuyt]

Forget to mention the strength of Damage which is the combination of an regressive deep search (often > 20 ply) in combination with the solid evaluation.
Although I need to improve a few cases for outposts and a few locks, which are not handled well.

Damage will become somewhere in the future available for public.
Need to think in what form.
Maybe ( = most likely) the GUI will still be free, but i will make a version available with a 8p DB and a engine on a mobile Hard-Disk, like Ed.
And the costs then will be mostly needed to cover for posting the version (and losing some in traffic ) and the costs of the HD.

For the time being it is still in improvement phase and getting better day by day.
I also will continue in posting match result and all kind of technical details for those interested.

Bert