lkaufman wrote:Two years ago FM John Meyer defeated Komodo by 3 to 1 (two wins, two draws) in a knight-odds match at 45' + 15". Two other players of roughly similar strength also scored 75% in eight games under those conditions. We have scheduled a rematch with the difference that this time the games will be chess960 (a.k.a. Fischerandom) with Komodo playing White but removing whichever knight is further from the center (or further from the king if otherwise tied). The human unfamiliarity with the start positions will surely favor Komodo, but whether it will do so by more than the 200 elo needed based on the previous match is not obvious. This will be the first public chess 960 handicap match on record as far as I know.
John Meyer has a FIDE rating of 2270, but it is not realistic as it is over a decade old. However his 2217 USCF rating is accurate, since he does play ragularly in USCF rated league matches. John is 72 years old, but still fairly healthy and in practice, and has remained consistently above his 2200 "floor", so he is definitely a legitimate master by US standards. Probably an accurate FIDE rating for him would be about a hundred lower, so 2117. He once tied for first in the US Open, and was probably strong enough to become an IM forty years ago if he had played internationally.
The match is scheduled for Thursday, June 22 at 11:30 a.m Easter Daylight Time with games every 2.5 to 3 hours. Four games total. Meyer will receive $30 per win and $10 per draw. Komodo will run on my 24 core xeon machine using the latest development versions. Broadcast from KomodoChess, I'll post the link before the match. If Komodo wins the match, I plan to try again against a stronger opponent.
3.0/4 for Komodo in my book.
I found a suitable adversary to test a bit against, Fruit 2.2.1, 900 ELO points weaker. I built ~100 positions Chess960 handicap book, and played with different Contempts. For some reason, a not very big contempt was favored. I was expecting Contempt=100 favored, it turned out 50 is better. Here are the result with Komodo being handicapped by a Knight against Fruit 2.2.1 (1000 games each run):
Also, Stockfish dev seems significantly stronger with this handicap against Fruit 2.2.1:
SF dev Contempt=0:
85.2% SF dev Contempt=50:
86.2%
SF dev Contempt=100:
85.1%
This suggests that the elo gap/15 might be a better guideline than the handicap value/3 for setting Contempt. Or perhaps higher Contempt is more appropriate against humans than against engines of the same strength.
What was the time control (very fast I suppose)? Do you have comparison data for knight odds at normal chess (with either 50 or 100 Contempt) under the same conditions? I think that Komodo will score significantly higher in chess960 knight odds even against an engine than it can in normal knight odds, even though familiarity and opening knowledge don't play a role. Many chess960 positions require some difficult decisions as to how best to develop and coordinate the pieces, I think.
Time control is very short indeed, 3''+0.03''.
Chess960 considerably dilates the rating differences (no odds matches):
Chess960
Contempt=50
Score of Komodo 11.01 vs Fruit 221: 1979 - 5 - 16 [0.994] 2000
ELO difference: 873.70 +/- 84.61
Finished match
3moves_GM
Contempt=50
Score of Komodo 11.01 vs Fruit 221: 1956 - 9 - 35 [0.987] 2000
ELO difference: 748.80 +/- 56.03
Finished match
I put on 3.0/4 for Komodo in this match, but I was tempted to put 3.5/4 too. The only countermeasure against Komodo in previous Knight-Odds matches was the opening preparation with forced lines. Now that is impossible, and besides that, as you noted, the strength gap now is larger than in normal chess Knight-Odds.
lkaufman wrote:Two years ago FM John Meyer defeated Komodo by 3 to 1 (two wins, two draws) in a knight-odds match at 45' + 15". Two other players of roughly similar strength also scored 75% in eight games under those conditions. We have scheduled a rematch with the difference that this time the games will be chess960 (a.k.a. Fischerandom) with Komodo playing White but removing whichever knight is further from the center (or further from the king if otherwise tied). The human unfamiliarity with the start positions will surely favor Komodo, but whether it will do so by more than the 200 elo needed based on the previous match is not obvious. This will be the first public chess 960 handicap match on record as far as I know.
John Meyer has a FIDE rating of 2270, but it is not realistic as it is over a decade old. However his 2217 USCF rating is accurate, since he does play ragularly in USCF rated league matches. John is 72 years old, but still fairly healthy and in practice, and has remained consistently above his 2200 "floor", so he is definitely a legitimate master by US standards. Probably an accurate FIDE rating for him would be about a hundred lower, so 2117. He once tied for first in the US Open, and was probably strong enough to become an IM forty years ago if he had played internationally.
The match is scheduled for Thursday, June 22 at 11:30 a.m Easter Daylight Time with games every 2.5 to 3 hours. Four games total. Meyer will receive $30 per win and $10 per draw. Komodo will run on my 24 core xeon machine using the latest development versions. Broadcast from KomodoChess, I'll post the link before the match. If Komodo wins the match, I plan to try again against a stronger opponent.
3.0/4 for Komodo in my book.
I found a suitable adversary to test a bit against, Fruit 2.2.1, 900 ELO points weaker. I built ~100 positions Chess960 handicap book, and played with different Contempts. For some reason, a not very big contempt was favored. I was expecting Contempt=100 favored, it turned out 50 is better. Here are the result with Komodo being handicapped by a Knight against Fruit 2.2.1 (1000 games each run):
Also, Stockfish dev seems significantly stronger with this handicap against Fruit 2.2.1:
SF dev Contempt=0:
85.2% SF dev Contempt=50:
86.2%
SF dev Contempt=100:
85.1%
This suggests that the elo gap/15 might be a better guideline than the handicap value/3 for setting Contempt. Or perhaps higher Contempt is more appropriate against humans than against engines of the same strength.
What was the time control (very fast I suppose)? Do you have comparison data for knight odds at normal chess (with either 50 or 100 Contempt) under the same conditions? I think that Komodo will score significantly higher in chess960 knight odds even against an engine than it can in normal knight odds, even though familiarity and opening knowledge don't play a role. Many chess960 positions require some difficult decisions as to how best to develop and coordinate the pieces, I think.
Time control is very short indeed, 3''+0.03''.
Chess960 considerably dilates the rating differences (no odds matches):
Chess960
Contempt=50
Score of Komodo 11.01 vs Fruit 221: 1979 - 5 - 16 [0.994] 2000
ELO difference: 873.70 +/- 84.61
Finished match
3moves_GM
Contempt=50
Score of Komodo 11.01 vs Fruit 221: 1956 - 9 - 35 [0.987] 2000
ELO difference: 748.80 +/- 56.03
Finished match
I put on 3.0/4 for Komodo in this match, but I was tempted to put 3.5/4 too. The only countermeasure against Komodo in previous Knight-Odds matches was the opening preparation with forced lines. Now that is impossible, and besides that, as you noted, the strength gap now is larger than in normal chess Knight-Odds.
At Knight-Odds the difference is very large too:
Chess960 Knight-Odds against Fruit Komodo Contempt=50:
82.0%
Normal Chess Knight-Odds against Fruit: Komodo Contempt=50:
65.0%
lkaufman wrote:Two years ago FM John Meyer defeated Komodo by 3 to 1 (two wins, two draws) in a knight-odds match at 45' + 15". Two other players of roughly similar strength also scored 75% in eight games under those conditions. We have scheduled a rematch with the difference that this time the games will be chess960 (a.k.a. Fischerandom) with Komodo playing White but removing whichever knight is further from the center (or further from the king if otherwise tied). The human unfamiliarity with the start positions will surely favor Komodo, but whether it will do so by more than the 200 elo needed based on the previous match is not obvious. This will be the first public chess 960 handicap match on record as far as I know.
John Meyer has a FIDE rating of 2270, but it is not realistic as it is over a decade old. However his 2217 USCF rating is accurate, since he does play ragularly in USCF rated league matches. John is 72 years old, but still fairly healthy and in practice, and has remained consistently above his 2200 "floor", so he is definitely a legitimate master by US standards. Probably an accurate FIDE rating for him would be about a hundred lower, so 2117. He once tied for first in the US Open, and was probably strong enough to become an IM forty years ago if he had played internationally.
The match is scheduled for Thursday, June 22 at 11:30 a.m Easter Daylight Time with games every 2.5 to 3 hours. Four games total. Meyer will receive $30 per win and $10 per draw. Komodo will run on my 24 core xeon machine using the latest development versions. Broadcast from KomodoChess, I'll post the link before the match. If Komodo wins the match, I plan to try again against a stronger opponent.
3.0/4 for Komodo in my book.
I found a suitable adversary to test a bit against, Fruit 2.2.1, 900 ELO points weaker. I built ~100 positions Chess960 handicap book, and played with different Contempts. For some reason, a not very big contempt was favored. I was expecting Contempt=100 favored, it turned out 50 is better. Here are the result with Komodo being handicapped by a Knight against Fruit 2.2.1 (1000 games each run):
Also, Stockfish dev seems significantly stronger with this handicap against Fruit 2.2.1:
SF dev Contempt=0:
85.2% SF dev Contempt=50:
86.2%
SF dev Contempt=100:
85.1%
This suggests that the elo gap/15 might be a better guideline than the handicap value/3 for setting Contempt. Or perhaps higher Contempt is more appropriate against humans than against engines of the same strength.
What was the time control (very fast I suppose)? Do you have comparison data for knight odds at normal chess (with either 50 or 100 Contempt) under the same conditions? I think that Komodo will score significantly higher in chess960 knight odds even against an engine than it can in normal knight odds, even though familiarity and opening knowledge don't play a role. Many chess960 positions require some difficult decisions as to how best to develop and coordinate the pieces, I think.
Time control is very short indeed, 3''+0.03''.
Chess960 considerably dilates the rating differences (no odds matches):
Chess960
Contempt=50
Score of Komodo 11.01 vs Fruit 221: 1979 - 5 - 16 [0.994] 2000
ELO difference: 873.70 +/- 84.61
Finished match
3moves_GM
Contempt=50
Score of Komodo 11.01 vs Fruit 221: 1956 - 9 - 35 [0.987] 2000
ELO difference: 748.80 +/- 56.03
Finished match
I put on 3.0/4 for Komodo in this match, but I was tempted to put 3.5/4 too. The only countermeasure against Komodo in previous Knight-Odds matches was the opening preparation with forced lines. Now that is impossible, and besides that, as you noted, the strength gap now is larger than in normal chess Knight-Odds.
At Knight-Odds the difference is very large too:
Chess960 Knight-Odds against Fruit Komodo Contempt=50:
82.0%
Normal Chess Knight-Odds against Fruit: Komodo Contempt=50:
65.0%
Based on your results, we can estimate that elo differences between engines increase by a bit under 20% playing chess 960. If this applies to the upcoming match, it would make Komodo a modest favorite. But if chess 960 is especially difficult for humans (compared to comparable - level engines), perhaps 3 to 1 would be a good guess. We'll see.
Modern Times wrote:Thanks for organising this match Larry ! Chess960 rocks
Since you are probably the person most familiar with chess960 on this site, are there any unusual situations that might come up that require more knowledge than just the basic rules of 960 castling? I was U.S. Open champion of chess 960 around 8 years ago and also gave a simul of chess 960, but I haven't played or watched more than perhaps fifty games so I'm no expert on the 960 positions. Let me know any subtle points I should explain to John Meyer in advance of the match, or if there are any ambiguities in the rules.
Modern Times wrote:Thanks for organising this match Larry ! Chess960 rocks
Since you are probably the person most familiar with chess960 on this site, are there any unusual situations that might come up that require more knowledge than just the basic rules of 960 castling? I was U.S. Open champion of chess 960 around 8 years ago and also gave a simul of chess 960, but I haven't played or watched more than perhaps fifty games so I'm no expert on the 960 positions. Let me know any subtle points I should explain to John Meyer in advance of the match, or if there are any ambiguities in the rules.
