Soccer-like chess

[mcostalba]

Firstly I congratulate with Komodo team for their nice win in TCEC 8: I guess for people was fun to watch, for SF fan a bit less, but that's the way it went, hopefully will be better next time.

With always more powerful engines and hardware we are witnessing an always increasing in draw rate. In TCEC8 superfinal around 90% of games were draws!

I don't think it depends on the ELO difference between opponents, even if Komodo or Stockfish play against a God engine, they will manage to draw most of their games, simply because they don't do enough mistakes to lose the game.

Even more surprising are the last 33 games, played with unbalanced opening. Here the engines immediately spotted opening was unbalanced, out of the book, but then in most of the cases, before move 30, black was able to equalize and even more surprising black won more times in these games than in the previous 66!

I think we have to get used to soccer-like final scores in these high level chess tournaments because this will be more and more common in the future...

[Norm Pollock]

Great match. Better luck next time. I'm sure your team will make many improvements- remove bugs, better time control, better evaluation and find a way to counter K's use of contempt = 0 (which encourages draws).

[lkaufman]

Firstly I congratulate with Komodo team for their nice win in TCEC 8: I guess for people was fun to watch, for SF fan a bit less, but that's the way it went, hopefully will be better next time.

With always more powerful engines and hardware we are witnessing an always increasing in draw rate. In TCEC8 superfinal around 90% of games were draws!

I don't think it depends on the ELO difference between opponents, even if Komodo or Stockfish play against a God engine, they will manage to draw most of their games, simply because they don't do enough mistakes to lose the game.

Even more surprising are the last 33 games, played with unbalanced opening. Here the engines immediately spotted opening was unbalanced, out of the book, but then in most of the cases, before move 30, black was able to equalize and even more surprising black won more times in these games than in the previous 66!

I think we have to get used to soccer-like final scores in these high level chess tournaments because this will be more and more common in the future...

Thank you. I don't agree with your opinion that Komodo or SF would draw most games against perfect play. It is easy enough to test this, if someone will devote the computer time needed; just play Komodo (or SF) against itself, with one getting TCEC time (or more if you have to use inferior hardware) and the other getting say four times as much time. If anyone does this please report results here.
There is a solution to the draw problem. There has to be some White advantage that will produce as many White wins as it does draws, plus some rare Black wins. So with sufficiently unbalanced openings, the draw percentage can be brought below 50% as well as the White win percentage. The only drawback is that this does not test the ability of the engines to create winning chances, only its ability to win marginally won games or to draw marginally drawn games.
Here is a perfect example: In the initial position, forbid Black from castling. Komodo gives that position a score that is around where I think the dividing line between win and draw lies. I had Komodo 9.3 play it out on four threads (to insure variety due to MP unpredictability) a hundred times at game in one minute. getting 49 White wins, 46 draws, 5 Black wins. Then at two minutes plus one second, a similar result: 49 White wins, 49 draws, 2 Black wins. So just forbidding castling for Black appears to put the game right on the win/draw line. Of course results could be different at TCEC time controls on many cores, but it's not obvious whether the White win percentage would go up or down.

[Laskos]

I don't think it depends on the ELO difference between opponents, even if Komodo or Stockfish play against a God engine, they will manage to draw most of their games, simply because they don't do enough mistakes to lose the game.

Well, I can show that Komodo (and probably Stockfish) will do better than God:

5-men chess is solved. For a better analogy with 32-men chess, which we assume to be draw in order for your argument to stand, I took 5-men drawn positions, several thousands of them, as starting positions. And a nice engine 1000 ELO points weaker than The God Komodo -- Zurichess. Ultra-fast games of Zurichess against Zurichess with 5-men theoretically drawn opening position show the following:

Code: Select all

Score of Zurichess vs Zurichess: 22 - 19 - 959  [0.501] 1000
ELO difference: 1
Finished match

The draw rate is 96%, your argument would be that 5-men chess is nearly solved by Zurichess, and "even if Zurichess plays against a God engine, it will manage to draw most of its games". But Zurichess has difficulty to draw most of its games from the same 5-men drawn positions against The God Komodo:

Code: Select all

Score of Komodo vs Zurichess: 538 - 3 - 459  [0.767] 1000
ELO difference: 207
Finished match

Only 46% draws.
I hope I have shown that Komodo is better than God. Maybe I will check Stockfish-God relationship too.

[hgm]

The 'unbalancing' of the openings was obviously done very poorly, and a complete bust. They were not anywhere near unbalanced enough. Pawn odds is known to produce around 20% advantage (i.e. a 30% score for the handicapped side), and even this falls short of the mark, as a 50-50 mix of draws and wins would cause a 75% score.

So it was predictable the out-of book advantage would have to be a Pawn and then something, and the chosen lines only provided a fraction of that.

[mcostalba]

your argument would be that 5-men chess is nearly solved by Zurichess

Actually I never said that: if I play from the same 5-men positions, I will get a score even worst than Zurichess...but if Komodo or SF plays from that positions on a superhardware and with a super TC than we will have a very long sequences of draws against _any_ God opponent.

[cdani]

Well, I can show that Komodo (and probably Stockfish) will do better than God:

Thanks for thinking and elaborating this good example! I also think, but is a guess, that we are far from have God-like engines. They draw most games thanks to having a lot of time but because they are tuned very similarly.

Also in your example, I expect Stockfish to behave mostly equal as Komodo, as I think that when there are less pieces the tactics are more relevant and the knowledge less.

[Laskos]

your argument would be that 5-men chess is nearly solved by Zurichess

Actually I never said that: if I play from the same 5-men positions, I will get a score even worst than Zurichess...but if Komodo or SF plays from that positions on a superhardware and with a super TC than we will have a very long sequences of draws against _any_ God opponent.

I was just saying that Komodo or SF on 32-men are probably no better than Zurichess on 5-men. Zurichess shows a 96% draw rate from 5-men against itself, and by your argument that makes Zurichess to draw most of the games from 5-men even against God. I have shown that God must be weaker than K in that case. In fact, it is likely that K or SF are worse on 32-men than Zurichess on 5-men, and the draw rate of K or SF on 32-men against God would be much lower than 46%, even having 90% draw rate between K and SF on 32-men in some conditions.

[Laskos]

Well, I can show that Komodo (and probably Stockfish) will do better than God:

Thanks for thinking and elaborating this good example! I also think, but is a guess, that we are far from have God-like engines. They draw most games thanks to having a lot of time but because they are tuned very similarly.

Also in your example, I expect Stockfish to behave mostly equal as Komodo, as I think that when there are less pieces the tactics are more relevant and the knowledge less.

Right, just tested:

Code: Select all

Score of Stockfish vs Zurichess: 523 - 9 - 468  [0.757] 1000
ELO difference: 197
Finished match

47% draws, and Stockfish is better than God too. I believe on 32-men K and SF will get significantly less draws against God than these 40+%, and that is also based on some more substantiated extrapolations either on plies or on doubling which Don Dailey, me and some others performed in the past (some posted here). This current apparent "draw death" is a bit artificial, as top engines are using similar techniques. In fact 50%-60% draw rates were occurring IIRC as far back as 2003 in long games of top engines, something which does not prevent them to score 2% against K and SF nowadays.

[mcostalba]

I think that when at around move 20 both engines report 0.00 draw score, there is nothing more to do.