Computer Chess Club
https://talkchess.com/
The NNs should be relatively weaker at this variant without having been trained on it, although I doubt they will be much weaker than normally. In my Komodo vs SF no castling Armageddon tests Komodo does better in no castling Armageddon than in normal chess, but Stockfish still wins, the elo gap is still too large for details like this to change the overall result. I think that the gap between humans and engines will be much larger with no castling. Maybe Komodo could even give knight odds to ordinary grandmasters in rapid in no castling chess!?Chessqueen wrote: ↑Fri Dec 13, 2019 10:34 pm Do you believe that it would be Komodo, Stockfish or maybe a version of NN like Fat Fritz or Leela ? https://www.chessbase.in/news/No-Castli ... g-variant-
Kramnik basically makes two good points. One is that the castling rule is illogical and arbitrary, and may have been introduced for reasons that make no sense today. The other is that the game is much more interesting when the players have to figure out for themselves how to get their kings to secure positions rather than have one simple rule that does it for them. Even though the no-castling version will quickly be deeply analyzed by engines, the lack of any trivial way to safeguard the king will insure that new moves will lead to truly different games and not just to transpositions or near-transpositions. I think that he is basically correct on both counts, although whether it is a better solution than FRC (960) is not clear to me. Both versions should lead to a fairly dramatic reduction in draws at high level, in my opinion.
Not sure about the dramatic reduction of draw rates. Engines for some reason at longer TC show only some 3-4% reduction in draw rate. Humans will have lower draw rates only before the opening theory is masteted, and on that Lc0 can be trained in 3 months to master well the openings.lkaufman wrote: ↑Sat Dec 14, 2019 7:05 pmKramnik basically makes two good points. One is that the castling rule is illogical and arbitrary, and may have been introduced for reasons that make no sense today. The other is that the game is much more interesting when the players have to figure out for themselves how to get their kings to secure positions rather than have one simple rule that does it for them. Even though the no-castling version will quickly be deeply analyzed by engines, the lack of any trivial way to safeguard the king will insure that new moves will lead to truly different games and not just to transpositions or near-transpositions. I think that he is basically correct on both counts, although whether it is a better solution than FRC (960) is not clear to me. Both versions should lead to a fairly dramatic reduction in draws at high level, in my opinion.
Kramnik argues (and I agree) that the reduction in human draw rates will be long-lasting, far beyond the time needed for engines to analyze the new startposition, because the variety of king positions will be much greater; it's not just a question of memorizing exact move sequences but of knowing how to play with a standard castled king setup. But anyway I'm glad you are getting promising results for NBSC, I feel it will be more acceptable to chess players than just NBC. With NBC I think 9 out of 10 GMs would choose White, but with NBSC I think the choice would be fairly even, which is critical for this to be accepted. At the very least, it should be preferred as a tiebreaker to the silly 5 vs 4 min Armageddon, it could be played at some reasonable time control like 15' + 10". If it becomes accepted that way, the next step is of course actual tournaments with it.Laskos wrote: ↑Sat Dec 14, 2019 8:29 pmNot sure about the dramatic reduction of draw rates. Engines for some reason at longer TC show only some 3-4% reduction in draw rate. Humans will have lower draw rates only before the opening theory is masteted, and on that Lc0 can be trained in 3 months to master well the openings.lkaufman wrote: ↑Sat Dec 14, 2019 7:05 pmKramnik basically makes two good points. One is that the castling rule is illogical and arbitrary, and may have been introduced for reasons that make no sense today. The other is that the game is much more interesting when the players have to figure out for themselves how to get their kings to secure positions rather than have one simple rule that does it for them. Even though the no-castling version will quickly be deeply analyzed by engines, the lack of any trivial way to safeguard the king will insure that new moves will lead to truly different games and not just to transpositions or near-transpositions. I think that he is basically correct on both counts, although whether it is a better solution than FRC (960) is not clear to me. Both versions should lead to a fairly dramatic reduction in draws at high level, in my opinion.
More interesting is NBC and NBSC, they are borderline, I have some results on NBSC, they are quite promising, I will post them later. The resolving power is amazing with these variants, which is one if the most important factors (playing fewer games for the same statisticsl significance). And the deviation from Chess is minimal.
I was talking about engines' playing, and it is known that they are not afraid of vulnerable Kings, they defend very accurately. And in their case, the draw rate is very mildly decreased. It is beyond my patzer human level to talk about how would a GM feel with an exposed King. Also, in my patzer view, GM are often all too eager to agree on a draw. Draws are maybe to avoid on this basis too.lkaufman wrote: ↑Sat Dec 14, 2019 9:05 pmKramnik argues (and I agree) that the reduction in human draw rates will be long-lasting, far beyond the time needed for engines to analyze the new startposition, because the variety of king positions will be much greater; it's not just a question of memorizing exact move sequences but of knowing how to play with a standard castled king setup. But anyway I'm glad you are getting promising results for NBSC, I feel it will be more acceptable to chess players than just NBC. With NBC I think 9 out of 10 GMs would choose White, but with NBSC I think the choice would be fairly even, which is critical for this to be accepted. At the very least, it should be preferred as a tiebreaker to the silly 5 vs 4 min Armageddon, it could be played at some reasonable time control like 15' + 10". If it becomes accepted that way, the next step is of course actual tournaments with it.Laskos wrote: ↑Sat Dec 14, 2019 8:29 pmNot sure about the dramatic reduction of draw rates. Engines for some reason at longer TC show only some 3-4% reduction in draw rate. Humans will have lower draw rates only before the opening theory is masteted, and on that Lc0 can be trained in 3 months to master well the openings.lkaufman wrote: ↑Sat Dec 14, 2019 7:05 pmKramnik basically makes two good points. One is that the castling rule is illogical and arbitrary, and may have been introduced for reasons that make no sense today. The other is that the game is much more interesting when the players have to figure out for themselves how to get their kings to secure positions rather than have one simple rule that does it for them. Even though the no-castling version will quickly be deeply analyzed by engines, the lack of any trivial way to safeguard the king will insure that new moves will lead to truly different games and not just to transpositions or near-transpositions. I think that he is basically correct on both counts, although whether it is a better solution than FRC (960) is not clear to me. Both versions should lead to a fairly dramatic reduction in draws at high level, in my opinion.
More interesting is NBC and NBSC, they are borderline, I have some results on NBSC, they are quite promising, I will post them later. The resolving power is amazing with these variants, which is one if the most important factors (playing fewer games for the same statisticsl significance). And the deviation from Chess is minimal.
GMs do often agree to draws prematurely, but there have been plenty of top level events with rules designed to prevent or minimize that, such as no draws by agreement before move 30 or 40, or none except by offer to TD who must approve before the offer is made to the opponent, or simply to stating that players won't be invited back if they agree to draws too early. All of these help a bit, but even with such rules the draw percentage is just unacceptably high in top level human play. In general, White does try to win the game, and usually gets a small edge, but not enough to win the game. Kramnik's rule would help a lot more in human play than in engine play, but it won't help much for correspondence play where engines are consulted or engine tournaments.Laskos wrote: ↑Sat Dec 14, 2019 9:54 pmI was talking about engines' playing, and it is known that they are not afraid of vulnerable Kings, they defend very accurately. And in their case, the draw rate is very mildly decreased. It is beyond my patzer human level to talk about how would a GM feel with an exposed King. Also, in my patzer view, GM are often all too eager to agree on a draw. Draws are maybe to avoid on this basis too.lkaufman wrote: ↑Sat Dec 14, 2019 9:05 pmKramnik argues (and I agree) that the reduction in human draw rates will be long-lasting, far beyond the time needed for engines to analyze the new startposition, because the variety of king positions will be much greater; it's not just a question of memorizing exact move sequences but of knowing how to play with a standard castled king setup. But anyway I'm glad you are getting promising results for NBSC, I feel it will be more acceptable to chess players than just NBC. With NBC I think 9 out of 10 GMs would choose White, but with NBSC I think the choice would be fairly even, which is critical for this to be accepted. At the very least, it should be preferred as a tiebreaker to the silly 5 vs 4 min Armageddon, it could be played at some reasonable time control like 15' + 10". If it becomes accepted that way, the next step is of course actual tournaments with it.Laskos wrote: ↑Sat Dec 14, 2019 8:29 pmNot sure about the dramatic reduction of draw rates. Engines for some reason at longer TC show only some 3-4% reduction in draw rate. Humans will have lower draw rates only before the opening theory is masteted, and on that Lc0 can be trained in 3 months to master well the openings.lkaufman wrote: ↑Sat Dec 14, 2019 7:05 pmKramnik basically makes two good points. One is that the castling rule is illogical and arbitrary, and may have been introduced for reasons that make no sense today. The other is that the game is much more interesting when the players have to figure out for themselves how to get their kings to secure positions rather than have one simple rule that does it for them. Even though the no-castling version will quickly be deeply analyzed by engines, the lack of any trivial way to safeguard the king will insure that new moves will lead to truly different games and not just to transpositions or near-transpositions. I think that he is basically correct on both counts, although whether it is a better solution than FRC (960) is not clear to me. Both versions should lead to a fairly dramatic reduction in draws at high level, in my opinion.
More interesting is NBC and NBSC, they are borderline, I have some results on NBSC, they are quite promising, I will post them later. The resolving power is amazing with these variants, which is one if the most important factors (playing fewer games for the same statisticsl significance). And the deviation from Chess is minimal.
Yes, NBSC seems very promising, and again there are 3 main cases: 1/ Komodo self-play (low White score, more draws), 2/ SF, Komodo and Houdini --- in RR, balanced Whire score (around 50%) and 3/ Lc0 different nets play, balanced too. Studying now to larger TC. What is even better is that the resolving power is even greater, and it seems to me that this happens because of another imbalance introduced - flank imbalance. White often preventively hits the Queen side and the game is complicated in decisions. For building hundreds of openings I used similarly to NBC Komodo on 4 threads with Contempt 75. I will post everything tomorrow.
Probably "NCC" because it seems Armageddon will be the last variant named by creative people
