Page 1 of 2

Knight odds match

Posted: Fri Sep 25, 2015 3:37 am
by lkaufman
Our next Komodo vs. human match is scheduled for this Monday, Sept. 28, at 11 a.m. Eastern time. Since we beat FM John Meyer decisively (3.5 to 0.5) at knight for pawn handicap, I offered him a rematch at full knight odds. Seven years ago he beat Rybka (near version 3) easily by four wins out of four games, so we'll see how much better (if any) Komodo can do now. So far Komodo has only scored 2 points out of 8 games in official match games, against an average opposition of 2162 FIDE, which is probably close to Meyer's current playing strength.
The details are: four games, 45' + 15" for Meyer, 30' + 10" for Komodo, Komodo always White but removing a knight, alternating between b1 and g1 knight for variety. Komodo will have a small handicap opening book. Meyer will receive $25 per win and just $5 per draw, so he won't be tempted to make draws by repetition while still winning.
Despite his drubbing at knight for pawn, Meyer expects to win at knight odds by at least 3.5 to 0.5. While I also expect him to win the match, I think it should be closer than the 25% we have scored so far at this handicap. My reasons:
1. Komodo mp has been improved perhaps 20 elo or so from the Bolzoni match.
2. We will have a small book this time, we didn't do that in the earlier matches.
3. Alternating b1 and g1 handicap should minimize the preparation/learning factor. I'll point out that although over 90% of knight odds played in the 1800s were b1 handicap, presumably because the handicap giver considered that to be the lesser handicap of the two, g1 handicap was also used and it was the handicap in the only recorded game in which Emanuel Lasker gave this handicap. I think that given the overwhelming preference to meet 1.e4 by e5 (which may have even been a condition for giving the handicap in some matches back then) b1 was the lesser handicap, but with modern day opening preferences and computer analysis the g1 handicap looks to me to be very slightly easier to give than the b1 handicap, at least not worse.
4. The Bolzoni match was played on a fast 6 core machine, whereas I'll be using my 24 core. The 3 to 2 time handicap we'll be giving negates only part of this difference.

I'll be very happy if we manage to tie the match. Meyer, despite his age (70), still plays in rated League play frequently, and still maintains a rating in the master (2200+) range without having hit the 2200 floor he gets as a Life Master. So if we do manage to draw the match, Komodo will have demonstrated that it can successfully give knight odds to a genuine chess master in a serious, "standard" level match. But that's a big "if".

Finally, I'll mention that our next scheduled match after this will be against GM Alex Lenderman at same handicaps as Movsesian. Details after the Meyer match.

Re: Knight odds match

Posted: Fri Sep 25, 2015 8:38 am
by thekingman
I'm going to optimistically guess 1.5/4. I don't doubt the strength of Komodo, and that's why I've guessed as high as I did, but still... a knight is a knight!

Re: Knight odds match

Posted: Fri Sep 25, 2015 9:07 am
by sainzlei
I predicted 3 Draw and 1 win for komodo :D

Re: Knight odds match

Posted: Fri Sep 25, 2015 12:23 pm
by Uri Blass
My opinion is that 45+15 is not what I consider to be serious chess and every serious tournament that I know is with slower time controls.

Even the Europe championship under age 8 that has clearly weaker level than this player is with time control of 90/40+30/game+30 seconds per move from the first move.

I also think that it is more interesting to see result in a tournament relative to result in a match against specific player.

Rating of humans is not based on their ability to win against computers with knight odd and it is possible that there are humans who are 200 elo stronger or 200 elo weaker than their rating when they play in these conditions.

If you find a tournament at the level of the following tournament then it is going to be more interesting from my point of view relative to a match only with one player when it is possible that the player is relatively strong or relatively weak against computers in the relevant conditions.

http://www.chess-results.com/tnr180030. ... =30&wi=984

Re: Knight odds match

Posted: Fri Sep 25, 2015 2:22 pm
by lkaufman
Uri Blass wrote:My opinion is that 45+15 is not what I consider to be serious chess and every serious tournament that I know is with slower time controls.

Even the Europe championship under age 8 that has clearly weaker level than this player is with time control of 90/40+30/game+30 seconds per move from the first move.

I also think that it is more interesting to see result in a tournament relative to result in a match against specific player.

Rating of humans is not based on their ability to win against computers with knight odd and it is possible that there are humans who are 200 elo stronger or 200 elo weaker than their rating when they play in these conditions.

If you find a tournament at the level of the following tournament then it is going to be more interesting from my point of view relative to a match only with one player when it is possible that the player is relatively strong or relatively weak against computers in the relevant conditions.

http://www.chess-results.com/tnr180030. ... =30&wi=984
A tournament might be more interesting, but it is not practical for me to run, although Jesse has some interest in the idea. Probably engines would get higher ratings in tournaments, since the opponents don't get to learn from experience and since they can't focus solely on handicap play. We have changed opponents for every match until this one, so the conclusions will be based on a range of opponents.

Regarding time control, I agree that the ratings we would get would be a bit lower at 90' + 30", although perhaps not by as much as you might think since fatigue/boredom may become a problem in a four hour game. But I would also say that people who buy Komodo and want to play against it with a handicap are much more likely to play two hour games than four hour games; probably one hour games would be most popular. I would estimate that humans play maybe 100 elo stronger at 90 + 30 than at 45 + 15 (less if you allow for fatigue), while the engine gains maybe half of that. So if perform at 2250 level (for example) giving knight odds at 45' + 15", we might expect to perform at just 2200 level at 90' + 30". The 45' + 15" level is the slowest one we can use and expect to get live coverage from chess.com for the GM matches.

Re: Knight odds match

Posted: Fri Sep 25, 2015 9:04 pm
by JJJ
I predict 2,5 for Komodo =)
I feel optimist about this one.

Re: Knight odds match

Posted: Fri Sep 25, 2015 9:31 pm
by velmarin
Hi larry.
We should make a section on bet. :twisted:

This is all just publicity of Komodo, which is a commercial product.
It has no other value.
Please!! :evil:

Re: Knight odds match

Posted: Fri Sep 25, 2015 9:36 pm
by Graham Banks
I'll go for 50%.

Re: Knight odds match

Posted: Fri Sep 25, 2015 10:40 pm
by lkaufman
velmarin wrote:Hi larry.
We should make a section on bet. :twisted:

This is all just publicity of Komodo, which is a commercial product.
It has no other value.
Please!! :evil:
The publicity is what justifies the cost of running these matches. I think it is of some general interest to learn whether the gap between the top engine and a real (but borderline) chess master has reached knight odds or not. It is also useful for people who want to play their own engine (whether Komodo or another one) at some handicap. I (like many others) found the Nakamura vs. Stockfish handicap match to be interesting too, and I encourage others to try such matches with other engines vs. human masters.

Re: Knight odds match

Posted: Fri Sep 25, 2015 11:02 pm
by velmarin
lkaufman wrote:
velmarin wrote:Hi larry.
We should make a section on bet. :twisted:

This is all just publicity of Komodo, which is a commercial product.
It has no other value.
Please!! :evil:
The publicity is what justifies the cost of running these matches. I think it is of some general interest to learn whether the gap between the top engine and a real (but borderline) chess master has reached knight odds or not. It is also useful for people who want to play their own engine (whether Komodo or another one) at some handicap. I (like many others) found the Nakamura vs. Stockfish handicap match to be interesting too, and I encourage others to try such matches with other engines vs. human masters.
Then the opponent of your engine,
¿do have a salary? ¿you pay that salary?