I agree that asymmetry may work because kingside castling is more common than queenside castling, but that is not the right way to do it.Lyudmil Tsvetkov wrote:An f6 pawn is the strongest pawn on the chess board, much stronger than e4,e5,d5,d6. And also much stronger than c6. Once I also believed asymmetric psqt is stupid, but when Adam posted some results of such psqt and I thougth about it afterwards, I came to the conclusion that some asymmetry in psqt is simply obligatory, if you want to be precise in assessing factors.arjuntemurnikar wrote:Of course SF will find the right path to attack if it is within its depth. But sometimes, it needs to search a bit deeper. In this case it is not a big surprise that it needs larger depths in the position you described. It's a closed position with many quiet moves possible and nothing is really obvious. SF finds Rh3 eventually, but it is not straightforward. Anyway I don't see how this can be improved much... There will always be some positions where depth is critical, due to the fact that inherently all engines use a brute-force depth-based algorithm. That's just how it is...Lyudmil Tsvetkov wrote:
Yeah, it is my slower hardware, SF rarely gets to depths bigger than 24 in blitz, I should buy a 64-core in the future.On the other hand, you know that some people test eval and search changes even at 15 and 60 sec. for the entire game.
King safety is both a matter of search and evaluation. On my machine SF never has problems with its attacking pieces, but it has problems with its attacking pawns.
The testing of patches at 15 and 60 secs has its drawbacks of course, but it works out in the end statistically because we play so many games.
Rest assured, I won't say there aren't any improvements to be made. There are always improvements possible...
For example, I would give bigger value to e4 than d4 central pawn. The much bigger likelihood that black castles short simply makes this pawn more relevant in attacking the king than the d4 pawn. e4 can go to e5 and then f6, while it is harder for d4 to do so. If you do not do this in psqt, difficult to do it in any other way. For me this is not asymmetric and ugly, we are just accustomed to symmetric tables.
Same with connected pawns: probability that files scores in SF should be not 1,3,3,4 from a to d but rather 1,2.5,3,4 is very high. And also probability that connected pawns should score higher on the kingside than the queenside, at least in the middlegame, as they influence there king attack, is also very high.
Just some random observations...
The right way would be to give different weights based on the castling side, rather than hardcode them into the psqt. Hardcoding is like putting a bandage on the problem rather than solving the problem itself. By doing that, you risk making tuning more difficult in the future for other parts of the eval because hardcoding asymmetry compromises orthogonality.
So asymmetric tables may bring short-term benefits, but in the long-term, it will hurt. That is why I am against it.