Tournament of programs in international draughts

[Krzysztof Grzelak]

Hi Krzysztof!

Great tournament!
If i may i will make a sugestion: in the next tournament you could make a "qualifying stage" between the previous last place (in this case Cerberus) and other programs like for example: Dream Dames, Plus 500, Windames 3D...
It's just an idea.

Please continue and good games.

Thank you for the idea. As for the program Dream Dames, I do not have the full version of this program - even though I wanted to buy it is not going to buy it, I have only a demo. The program Plus 500 have a full version with a license, while the program Windames 3D program is difficult to set a time of 20 minutes for 75 moves. I wanted to write to the author program Windames 3D, but unfortunately I do not have email.

Krzysztof.

[Jelle Wiersma]

Interesting top four!
Apparently Truus is still one of the best.

[MichelG]

Interesting top four!
Apparently Truus is still one of the best.

Actually, i think it is a bit depressing that the modern programs still can't reliably win from a 30 year old one that has no multithreading, no bitboards and no machine learning.

I think that soon I will do a new tournament among programs 18 rounds system round-robin.

You could always reduce the time per game if you want to run more rounds.

[Rein Halbersma]

Interesting top four!
Apparently Truus is still one of the best.

Actually, i think it is a bit depressing that the modern programs still can't reliably win from a 30 year old one that has no multithreading, no bitboards and no machine learning.

I think that Damage and Kingsrow (and your program probably as well) have shown to reliably beat Truus/Flits over 158-game matches. For a 7 or 9 round tournament, the number of games is simply too small to overcome the high drawing margin in draughts.

[Krzysztof Grzelak]

You could always reduce the time per game if you want to run more rounds.

I think that I would stay at 20 minutes 75 moves. Only tournament I played great and I played two rounds a day. In this case, the tournament will last for 9 days. My dream is to see such a tournament live broadcast over the Internet. Unfortunately, I can not afford to buy the electronic board DGT board and transmit batches over the internet.

I think that Damage and Kingsrow (and your program probably as well) have shown to reliably beat Truus/Flits over 158-game matches. For a 7 or 9 round tournament, the number of games is simply too small to overcome the high drawing margin in draughts.

I think you're right, but I write only about the program Kingsrow. Also note that no tournament is not 158 rounds.

[Ed Gilbert]

Hi Krzysztof,

I wonder what your goal or purpose is in running these tournaments. Is it primarily to experiment with the various programs and have some fun, or is it to determine the relative strengths of the programs? To get any kind of accurate comparisons of their relative strengths, it takes a lot more games. If programs are close in strength, it takes hundreds, sometimes thousands of games, to get any useful data. The only practical way to do it is to use faster games, and automate them. When I make a change to kingsrow and want to see if it made an improvement, I usually play ~8000 games, and sometimes that is not enough to know the answer with good confidence.

-- Ed

[Jelle Wiersma]

In my opinion playing 75 moves in 20 minutes is OK.
Actually I would even recommend 30 minutes per game and two games (1x white, 1x black).
In my opinion more time means emphasis on positional play, less time means emphasis on speed.
Regarding Truus, I believe that his program was designed for positional play, not for blitz...

[MichelG]

In my opinion playing 75 moves in 20 minutes is OK.
Actually I would even recommend 30 minutes per game and two games (1x white, 1x black).
In my opinion more time means emphasis on positional play, less time means emphasis on speed.
Regarding Truus, I believe that his program was designed for positional play, not for blitz...

In draughts, there are no forced combinations beyond about 6-7 ply (not counting captures), so positional play already dominates on short time settings.

The problem with such a long game-time, is that the outcome of a game, or short match won't tell you anything about which program or change is better. As Ed pointed out; to confirm small improvements you need thousands of game results. In general, if you have a version A that is better then version B, it will be so regardless of the time settings; the win/lose ratio stays the same, only the draw rate increases.

I think the difference in viewpoint is determined by what you want to do; either you use lots of games to see if a change improves the program, or you just play a few and try to analyse them individually by hand to determine if you can find the cause of weaknesses. Both ideas are useful.

Michel

[Krzysztof Grzelak]

Some time ago, one of my colleagues asked me this question. Krzysztof how would you like to have time for a game during the tournament program. And I answered very quickly - 1,20h + 1 min/move.

[tagliani45]

Let me say that with 4 proggies scoring 10 points each (4 winners in my view) the result seems stupid.Why not simply applying humans' tournaments rules.The result would be more significative.
That's the only way ,even if it takes a lot of time ,to end up with an reliable ranking.

[Krzysztof Grzelak]

Let me say that with 4 proggies scoring 10 points each (4 winners in my view) the result seems stupid.Why not simply applying humans' tournaments rules.The result would be more significative.
That's the only way ,even if it takes a lot of time ,to end up with an reliable ranking.

I think that how I will make new tournament this I will apply Sonneborn-Berger score.

Krzysztof.