Stockfish Handicap Matches

[lkaufman]

Would he be allowed to play the game himself on chess.com while commenting on his own site? That would avoid a lot of problems, including the time delay for me to observe and make his move, the chance of a mistake by me, and the issue of having to play as myself while actually making his moves? There probably wouldn't be a chess.com show involved, it would just be a game he played on chess.com and chose to comment on on his own. That is not a problem for chess.com or Komodo, but it might not be permitted by his own situation, I wouldn't know. If it is, all I need is an agreement on what time to send the challenge, and he just needs to accept. Same for any other reasonably strong player (let's say over 2100 FIDE) who would like to try to play Komodo on chess.com at knight odds, we just need to agree on the conditions of the game and the time to start.

[Chessqueen]

Would he be allowed to play the game himself on chess.com while commenting on his own site? That would avoid a lot of problems, including the time delay for me to observe and make his move, the chance of a mistake by me, and the issue of having to play as myself while actually making his moves? There probably wouldn't be a chess.com show involved, it would just be a game he played on chess.com and chose to comment on on his own. That is not a problem for chess.com or Komodo, but it might not be permitted by his own situation, I wouldn't know. If it is, all I need is an agreement on what time to send the challenge, and he just needs to accept.

Same for any other reasonably strong player (let's say over 2100 FIDE) who would like to try to play Komodo on chess.com at knight odds, we just need to agree on the conditions of the game and the time to start.

You mentioned players over 2100 FIDE but up to what rating with knight odds? Since GM David Smerdon Rated 2508 FIDE was a little bit too much, there are a few friends who might be interested if the TC is at least 45 + 5 increments but some of them are close to 2375 FIDE. But if there are between 2375 to 2550 what would be the odds that you are willing to give them at 30 + 10 seconds increment ?

[lkaufman]

Would he be allowed to play the game himself on chess.com while commenting on his own site? That would avoid a lot of problems, including the time delay for me to observe and make his move, the chance of a mistake by me, and the issue of having to play as myself while actually making his moves? There probably wouldn't be a chess.com show involved, it would just be a game he played on chess.com and chose to comment on on his own. That is not a problem for chess.com or Komodo, but it might not be permitted by his own situation, I wouldn't know. If it is, all I need is an agreement on what time to send the challenge, and he just needs to accept.

Same for any other reasonably strong player (let's say over 2100 FIDE) who would like to try to play Komodo on chess.com at knight odds, we just need to agree on the conditions of the game and the time to start.

You mentioned players over 2100 FIDE but up to what rating with knight odds? Since GM David Smerdon Rated 2508 FIDE was a little bit too much, there are a few friends who might be interested if the TC is at least 45 + 5 increments but some of them are close to 2375 FIDE. But if there are between 2375 to 2550 what would be the odds that you are willing to give them at 30 + 10 seconds increment ?

At 30 +10 I would estimate that knight odds is about fair for 2250 FIDE. Armageddon knight odds (meaning knight odds but Black loses draws) might be fair around 2350 or so. The next handicap below knight odds classically is "pawn and three moves", meaning remove f7 pawn, White plays 1e4 and 2d4, and it is still White's turn to make the third move. An alternative that I like which was not used in the classical period is c7 + f7 odds, which is the largest possible two pawn handicap except for the almost ridiculous f7 + g7, where Black can be forced to move his king in the opening. Pawn and 3 and c7/f7 are about equally large (the "c" pawn is worth about two tempi), and either handicap is suitable for the rating range you mention in my opinion. They are each something like a 2.5 pawn handicap or so. I'll be fine with either handicap for that rating range at 30' + 10"; below 2400 we could do Knight odds armageddon, although Komodo would be the underdog in that case. I would just need the player's name, rating(s), chess.com handle, number of games to be played, handicap(s), and agreement as to date and time. When there's prize money involved and a show I need to know that the player has a good reputation and won't cheat, but I think it's safe to assume that anyone playing just for the challenge without a show or money involved will not cheat. Note that these are significantly larger handicaps than Stockfish gave to van Foreest at that time limit, the difference of playing White vs. Black with f7 missing is huge. For opponents above the 2550 rating you mention, then Komodo probably needs to play White (c2/f2).

[chrisw]

Would he be allowed to play the game himself on chess.com while commenting on his own site? That would avoid a lot of problems, including the time delay for me to observe and make his move, the chance of a mistake by me, and the issue of having to play as myself while actually making his moves? There probably wouldn't be a chess.com show involved, it would just be a game he played on chess.com and chose to comment on on his own. That is not a problem for chess.com or Komodo, but it might not be permitted by his own situation, I wouldn't know. If it is, all I need is an agreement on what time to send the challenge, and he just needs to accept.

Same for any other reasonably strong player (let's say over 2100 FIDE) who would like to try to play Komodo on chess.com at knight odds, we just need to agree on the conditions of the game and the time to start.

You mentioned players over 2100 FIDE but up to what rating with knight odds? Since GM David Smerdon Rated 2508 FIDE was a little bit too much, there are a few friends who might be interested if the TC is at least 45 + 5 increments but some of them are close to 2375 FIDE. But if there are between 2375 to 2550 what would be the odds that you are willing to give them at 30 + 10 seconds increment ?

At 30 +10 I would estimate that knight odds is about fair for 2250 FIDE. Armageddon knight odds (meaning knight odds but Black loses draws) might be fair around 2350 or so. The next handicap below knight odds classically is "pawn and three moves", meaning remove f7 pawn, White plays 1e4 and 2d4, and it is still White's turn to make the third move. An alternative that I like which was not used in the classical period is c7 + f7 odds, which is the largest possible two pawn handicap except for the almost ridiculous f7 + g7, where Black can be forced to move his king in the opening. Pawn and 3 and c7/f7 are about equally large (the "c" pawn is worth about two tempi), and either handicap is suitable for the rating range you mention in my opinion. They are each something like a 2.5 pawn handicap or so. I'll be fine with either handicap for that rating range at 30' + 10"; below 2400 we could do Knight odds armageddon, although Komodo would be the underdog in that case. I would just need the player's name, rating(s), chess.com handle, number of games to be played, handicap(s), and agreement as to date and time. When there's prize money involved and a show I need to know that the player has a good reputation and won't cheat, but I think it's safe to assume that anyone playing just for the challenge without a show or money involved will not cheat. Note that these are significantly larger handicaps than Stockfish gave to van Foreest at that time limit, the difference of playing White vs. Black with f7 missing is huge. For opponents above the 2550 rating you mention, then Komodo probably needs to play White (c2/f2).

The pawn odds epds on my github page are broken (somehow, managed to remove the h-pawn rather than the f-pawn) but can be rebuilt if they are needed. As can the other variants. Let me know if it would be useful to fix and regenerate.

[Rebel]

Made a summary of the knight-odds discussion, includes the download of 6000 games.

http://rebel13.nl/rebel13/experiments.html#11

[lkaufman]

Would he be allowed to play the game himself on chess.com while commenting on his own site? That would avoid a lot of problems, including the time delay for me to observe and make his move, the chance of a mistake by me, and the issue of having to play as myself while actually making his moves? There probably wouldn't be a chess.com show involved, it would just be a game he played on chess.com and chose to comment on on his own. That is not a problem for chess.com or Komodo, but it might not be permitted by his own situation, I wouldn't know. If it is, all I need is an agreement on what time to send the challenge, and he just needs to accept.

Same for any other reasonably strong player (let's say over 2100 FIDE) who would like to try to play Komodo on chess.com at knight odds, we just need to agree on the conditions of the game and the time to start.

You mentioned players over 2100 FIDE but up to what rating with knight odds? Since GM David Smerdon Rated 2508 FIDE was a little bit too much, there are a few friends who might be interested if the TC is at least 45 + 5 increments but some of them are close to 2375 FIDE. But if there are between 2375 to 2550 what would be the odds that you are willing to give them at 30 + 10 seconds increment ?

At 30 +10 I would estimate that knight odds is about fair for 2250 FIDE. Armageddon knight odds (meaning knight odds but Black loses draws) might be fair around 2350 or so. The next handicap below knight odds classically is "pawn and three moves", meaning remove f7 pawn, White plays 1e4 and 2d4, and it is still White's turn to make the third move. An alternative that I like which was not used in the classical period is c7 + f7 odds, which is the largest possible two pawn handicap except for the almost ridiculous f7 + g7, where Black can be forced to move his king in the opening. Pawn and 3 and c7/f7 are about equally large (the "c" pawn is worth about two tempi), and either handicap is suitable for the rating range you mention in my opinion. They are each something like a 2.5 pawn handicap or so. I'll be fine with either handicap for that rating range at 30' + 10"; below 2400 we could do Knight odds armageddon, although Komodo would be the underdog in that case. I would just need the player's name, rating(s), chess.com handle, number of games to be played, handicap(s), and agreement as to date and time. When there's prize money involved and a show I need to know that the player has a good reputation and won't cheat, but I think it's safe to assume that anyone playing just for the challenge without a show or money involved will not cheat. Note that these are significantly larger handicaps than Stockfish gave to van Foreest at that time limit, the difference of playing White vs. Black with f7 missing is huge. For opponents above the 2550 rating you mention, then Komodo probably needs to play White (c2/f2).

The pawn odds epds on my github page are broken (somehow, managed to remove the h-pawn rather than the f-pawn) but can be rebuilt if they are needed. As can the other variants. Let me know if it would be useful to fix and regenerate.

f2 handicap has been used on some CCC engine tournaments, so might be useful, though it's too small for humans vs. computers. I think that the most interesting two pawn handicap is c2/f2, if you want to do a two pawn handicap database. That is likely to appear in future computer vs. top GM matches.

[lkaufman]

Made a summary of the knight-odds discussion, includes the download of 6000 games.

http://rebel13.nl/rebel13/experiments.html#11

Regarding your Benjamin program playing Komodo 14 with knight odds, final score at 2' + 1" (Komodo 7 threads, Benjamin 1 thread) was 11 to 9 in favor of Benjamin, +35 elo. For comparison, I ran same match K14 vs Fruit 2.2.1 same conditions, and Komodo won 11.5 to 8.5 (+53 elo). Since Fruit 2.2.1 is rated about 106 elo above Benjamin on CCRL Blitz list, Benjamin performed nearly 200 elo above Fruit relative to their CCRL blitz ratings. Quite a large difference!
I also ran a recent Lc0 net (70380) vs. Benjamin at knight odds same way (Lc0 used 2080 GPU + 2 threads cpu). This is the Lc0 network I think is best at giving knight odds. It lost 6.5 - 13.5, well below the 9 points scored by Komodo 14. So given your other results (K14 25 elo above SF11 at giving knight odds), Komodo 14 looks to be the best engine at giving knight odds, and Benjamin looks to be the best engine in its rating range at taking knight odds.

[Chessqueen]

Made a summary of the knight-odds discussion, includes the download of 6000 games.

http://rebel13.nl/rebel13/experiments.html#11

As I mentioned before in a previous post engines above 2700 SSDF are overrated compared to Human, therefore, when you concluded that the turning point lies around ~~2700 elo for engines, we can not conclude that most GM's around ~~2700 elo will lose against Stockfish with a knight Odds. As you will notice that a human rated around 2500 FIDE like GM Ben Finegold knows how to take advantage better than any engine rated around ~~2700 by the SSDF since those engines do not know how to force the trade of other pieces when winning bu a Knight or Bishop. Here I will show you what GM Ben Finegold was saying the day of GM Smerdon vs Komodo Knight Odds ==>

PS: It also look like GM Ben was making fun of Stockfish, so he should be the next match against Komodo with the (c2/f2) pawns Odds, and see if he can show us how to beat Komodo. Also Noticed that GM Ben Finegold made more Money $$$.$$ than GM Smerdon receiving Donations just to make commentaries

[xr_a_y]

Can we ask Lichess crew to let some engines with pawn, knight, ... odds (depending on their rating) play against humans ?
Some hours of such a tournament will probably give some insight on the subject ?

[Rebel]

Made a summary of the knight-odds discussion, includes the download of 6000 games.

http://rebel13.nl/rebel13/experiments.html#11

Regarding your Benjamin program playing Komodo 14 with knight odds, final score at 2' + 1" (Komodo 7 threads, Benjamin 1 thread) was 11 to 9 in favor of Benjamin, +35 elo. For comparison, I ran same match K14 vs Fruit 2.2.1 same conditions, and Komodo won 11.5 to 8.5 (+53 elo). Since Fruit 2.2.1 is rated about 106 elo above Benjamin on CCRL Blitz list, Benjamin performed nearly 200 elo above Fruit relative to their CCRL blitz ratings. Quite a large difference!
I also ran a recent Lc0 net (70380) vs. Benjamin at knight odds same way (Lc0 used 2080 GPU + 2 threads cpu). This is the Lc0 network I think is best at giving knight odds. It lost 6.5 - 13.5, well below the 9 points scored by Komodo 14. So given your other results (K14 25 elo above SF11 at giving knight odds), Komodo 14 looks to be the best engine at giving knight odds, and Benjamin looks to be the best engine in its rating range at taking knight odds.

Does a website exists that has recorded the Komodo - GM handicap matches?