Moderator: Ras
deletecorres wrote:You are a 1.c4 fan, as we know.Lyudmil Tsvetkov wrote:
it might be the case that 1.c4 is winning altogether, at least in some 90% of lines.
Nowadays 1.c4 is not a modish opening and this is the only one advantage for it.
corres wrote: and last /quote]
Sure. More code is requiredcorres wrote:You are right, Stockfish has a lot of weak sides.cdani wrote:
But from time to time I feel the need to say that Stockfish is very far from perfection. For example many times wining an endgame requires to do non standard things, so general tuning will prevent the engine to see such win, specially from a very previous position.
But this is not a tuning issue.
These cause the lack of the appropriate logic from the source code.
However writing this missing logic into the code this yields a more slower and generally more weaker Stockfish.
I simplified the phrase.Code: Select all
Andscacs 0.91:
info depth 26 seldepth 42 score cp 289 nodes 14405850 nps 1523783 tbhits 0 time
9454 pv e5d4 b6d6 d5e6 f5e6 d4a7 e6d7 a7c5 d6d5 b5b6 d7c6 g1h2 g8f7 c5a3 c6b7 f2
g3 d5f5 a3b2 f7e8 b2c3 f5b1 c3d4 b1e4 d4e5 e4c2 g3g4 b7d5 g4g5 c2f5 g5g3
info depth 26 currmove g2g4 currmovenumber 3
info depth 26 currmove e5f6 currmovenumber 7
info depth 27 currmove e5d4 currmovenumber 1
info depth 27 seldepth 51 score cp 305 lowerbound nodes 17548918 nps 1519913 tbh
its 0 time 11546 pv e5d4
info depth 27 currmove e5d4 currmovenumber 1
info depth 27 seldepth 59 score cp 321 lowerbound nodes 35732492 nps 1511398 tbh
its 0 time 23642 pv e5d4
info depth 27 currmove e5d4 currmovenumber 1
info depth 27 currmove g2g4 currmovenumber 2
info depth 27 seldepth 65 score cp 345 lowerbound nodes 56276154 nps 1506764 tbh
its 0 time 37349 pv g2g4
info depth 27 currmove g2g4 currmovenumber 1
info depth 27 seldepth 45 score cp 381 lowerbound nodes 59066754 nps 1512128 tbh
its 0 time 39062 pv g2g4
info depth 27 currmove g2g4 currmovenumber 1
info depth 27 seldepth 51 score cp 435 lowerbound nodes 61824257 nps 1517681 tbh
its 0 time 40736 pv g2g4
info depth 27 currmove g2g4 currmovenumber 1
info depth 27 seldepth 54 score cp 516 lowerbound nodes 66230815 nps 1527146 tbh
its 0 time 43369 pv g2g4
info depth 27 currmove g2g4 currmovenumber 1
info depth 27 seldepth 65 score cp 637 lowerbound nodes 71890476 nps 1542780 tbh
its 0 time 46598 pv g2g4
SF fails just because of too aggressive pruning, it has nothing to do with opposite coloured bishops.cdani wrote:Sure. More code is required
Example:
[d]6k1/p7/1q2p1p1/1P1PBb2/5P2/7P/5QP1/6K1 w - - 0 42
Stockfish and Komodo struggle to find a win as sure other engines, and Andscacs can find one relatively fast. The key is to exchange the queens plus doing the g4 move, and as most engines have a big plain penalization for different color bishop endgames, they don't realize it.
Stockfish don't find it, at least with the time I tried.Code: Select all
Andscacs 0.91: info depth 26 seldepth 42 score cp 289 nodes 14405850 nps 1523783 tbhits 0 time 9454 pv e5d4 b6d6 d5e6 f5e6 d4a7 e6d7 a7c5 d6d5 b5b6 d7c6 g1h2 g8f7 c5a3 c6b7 f2 g3 d5f5 a3b2 f7e8 b2c3 f5b1 c3d4 b1e4 d4e5 e4c2 g3g4 b7d5 g4g5 c2f5 g5g3 info depth 26 currmove g2g4 currmovenumber 3 info depth 26 currmove e5f6 currmovenumber 7 info depth 27 currmove e5d4 currmovenumber 1 info depth 27 seldepth 51 score cp 305 lowerbound nodes 17548918 nps 1519913 tbh its 0 time 11546 pv e5d4 info depth 27 currmove e5d4 currmovenumber 1 info depth 27 seldepth 59 score cp 321 lowerbound nodes 35732492 nps 1511398 tbh its 0 time 23642 pv e5d4 info depth 27 currmove e5d4 currmovenumber 1 info depth 27 currmove g2g4 currmovenumber 2 info depth 27 seldepth 65 score cp 345 lowerbound nodes 56276154 nps 1506764 tbh its 0 time 37349 pv g2g4 info depth 27 currmove g2g4 currmovenumber 1 info depth 27 seldepth 45 score cp 381 lowerbound nodes 59066754 nps 1512128 tbh its 0 time 39062 pv g2g4 info depth 27 currmove g2g4 currmovenumber 1 info depth 27 seldepth 51 score cp 435 lowerbound nodes 61824257 nps 1517681 tbh its 0 time 40736 pv g2g4 info depth 27 currmove g2g4 currmovenumber 1 info depth 27 seldepth 54 score cp 516 lowerbound nodes 66230815 nps 1527146 tbh its 0 time 43369 pv g2g4 info depth 27 currmove g2g4 currmovenumber 1 info depth 27 seldepth 65 score cp 637 lowerbound nodes 71890476 nps 1542780 tbh its 0 time 46598 pv g2g4
Different color bishop endgames require more code that a single penalization, as many times can be won and nobody notices it.
. It has to do with opposite coloured bishops because the pruned moves are the ones with less eval. And Andscacs gives more eval to this bishop endgame due to better eval coding of this endgame.Milos wrote: SF fails just because of too aggressive pruning, it has nothing to do with opposite coloured bishops.
http://www.andscacs.comWell if you disable LMR SF also finds it quickly without any modification of eval. So it's the way how you look at it.cdani wrote:It has to do with opposite coloured bishops because the pruned moves are the ones with less eval. And Andscacs gives more eval to this bishop endgame due to better eval coding of this endgame.Milos wrote: SF fails just because of too aggressive pruning, it has nothing to do with opposite coloured bishops.
If I remove the extra eval code of Andscacs or I tune it different, it also does not find the idea.
The engines will always aggressively prune, as they have no other way to go deep enough.
Of course, but my solution increases the strength of the engine (the added code was a win), and disabling lmr decreases it.Milos wrote: Well if you disable LMR SF also finds it quickly without any modification of eval. So it's the way how you look at it.
http://www.andscacs.com http://www.andscacs.com