Although, people did mention a minor piece, though I personally believe that would be too much for the program to give up. I think the exchange would be more competitive and allow for a close match.
If I'm remembering correctly, those games were closely contested.
Of course, I am a neophyte at computer chess but are computer ratings worthless. If a human was rated 3000+ could he not spot another human (2274) a knight?
Mike S. wrote:FM John Meyer with 2284 Elo won all four knight odds games against Rybka on an octo, in a 30+30 match.
No draws
This shows that engines can get outbooked by humans.
It also highlights the importance of book guidance for any engines.
I encourage some of you to try knight odds games. It is a huge advantage and even I can often win and draw rather easy most games. As long as you have sound endgame play the game needs only to force exchanges to enter a favourable endgame. This does not imply you cannot lose, I have done it and lost badly, however, the chance to win is there. I'll admit it is not something that interests me these days, but it can be alot of fun.
I think Rybka could be designed to play much stronger than it did with Knight odds. The problem is that it sees itself totally lost and doesn't see a way out. If the algo would try and complicate the position even if it is worse then the outcome might be different. For example it should be "taught" to stay in positions with as many pieces on the board as possible and to avoid any static non dynamic positions. The standard algo for normal chess simply is not tuned for that type of play. Still ... a knight handicap is a pretty big handicap and any reasonable chess player with some ability will simply try to exchange pieces and go into a dry technical ending a piece up.
Last edited by M ANSARI on Tue Jan 26, 2010 2:20 pm, edited 1 time in total.
Although, people did mention a minor piece, though I personally believe that would be too much for the program to give up. I think the exchange would be more competitive and allow for a close match.
If I'm remembering correctly, those games were closely contested.
Of course. With an extra piece, I feel confident I can win any game against anybody. An exchange is different. The key is how you convert an advantage. Converting a piece into a win is easy. Just not blundering, is enough. No matter what position you are, a piece is a piece. In many positions, a rook is not better than a knight or bishop. To win with a rook, you have to open lines, and we know what happens when you do that against a computer and you are not careful...
james uselton wrote:Of course, I am a neophyte at computer chess but are computer ratings worthless. If a human was rated 3000+ could he not spot another human (2274) a knight?
I think this shows that ELO difference is not a real mesure of outgoing result in an odd game.
A 2274 player surely is a FM, which is a chess player that know enough theory and have enough practice and enough pattern recognition to be able to hold (or try to win against) a 2800 human (Carlsen) or 3000 computer (Rybka) player wiht knight odds.
But a 1500 player surely will lose against a 2200 player (FM) with knight odds. He can be chased quite easy by setting a good trap, or just doesn't know how to force favourable exchanges to liquidate material.
