That is not close to blocked.mcostalba wrote:Ok I have pushed a branch called closed_positions:mcostalba wrote: I can write a closed position detector to be used by people to write tests on it
https://github.com/mcostalba/Stockfish/ ... d_position
with the code that implements your definition. Now I am able to detect positions like this one:
[d]r1q2rk1/2p3pp/2Pp4/p2P1p2/Q3p3/4P3/PP1R2PP/1K5R b - - 0 23
Tuning
Moderator: Ras
-
bob
- Posts: 20943
- Joined: Mon Feb 27, 2006 7:30 pm
- Location: Birmingham, AL
Re: Huge simplification
-
AdminX
- Posts: 6363
- Joined: Mon Mar 13, 2006 2:34 pm
- Location: Acworth, GA
Re: Huge simplification
That's semi-open!bob wrote:That is not close to blocked.mcostalba wrote:Ok I have pushed a branch called closed_positions:mcostalba wrote: I can write a closed position detector to be used by people to write tests on it
https://github.com/mcostalba/Stockfish/ ... d_position
with the code that implements your definition. Now I am able to detect positions like this one:
[d]r1q2rk1/2p3pp/2Pp4/p2P1p2/Q3p3/4P3/PP1R2PP/1K5R b - - 0 23
"Good decisions come from experience, and experience comes from bad decisions."
__________________________________________________________________
Ted Summers
__________________________________________________________________
Ted Summers
-
mcostalba
- Posts: 2684
- Joined: Sat Jun 14, 2008 9:17 pm
Re: Huge simplification
I have generalized the close position detector to evaluate any chain of blocked pawns: now is very general and flexible, the chain is evaluated according to its size and if is in the center or not. I have submitted some tests although very probably weights are wrong so I don't expect nice surprises.
-
arjuntemurnikar
- Posts: 204
- Joined: Tue Oct 15, 2013 10:22 pm
- Location: Singapore
Re: Huge simplification
Yes same thing. 5 ranks are included: Assuming King was RANK_1, then 1, 2, 3, 4, and 5 are included.Lyudmil Tsvetkov wrote:Hi Arjun.
By 4th rank I meant considering in storm danger f4,g4, h4 pawns as dangerous for a black king on g8.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Re: Huge simplification
Because, in some particular types of position, SF frequently does not see pushes to the 4th rank like f4,g4 and h4, especially when both kings have castled on the same side.arjuntemurnikar wrote:Yes same thing. 5 ranks are included: Assuming King was RANK_1, then 1, 2, 3, 4, and 5 are included.Lyudmil Tsvetkov wrote:Hi Arjun.
By 4th rank I meant considering in storm danger f4,g4, h4 pawns as dangerous for a black king on g8.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Re: Why this does not work?
2 further games where SF underestimates doubled rooks on an open file.
1r6/4r1k1/1pbp2p1/q1p4p/P1Pp1Pn1/3P1BP1/1R1N3P/RQ4K1 b - - 0 61
SF has just played 61.Ne4-d2 with a zero score.
4r3/4r1k1/1p1p2p1/q1p4p/P1Pp1Pn1/3P1NP1/1R5P/RQ4K1 w - - 0 63
After Gull doubles its rooks along the e file, it sees +70cps black edge. SF here still thinks white is worse by only some 30-40cps.
q3r3/6k1/1R1p2p1/2p4p/P1Pp1PnN/3P2P1/4r2P/RQ4K1 w - - 0 65
Upon entering the 7th rank, the black rook decides the game. Doubled rooks on an open file are so forceful, because they remain unchallenged.
2r2rk1/pb2q1pp/1pn1p3/3p4/3Pnp2/PP1Q1NP1/1BR1PPBP/2R3K1 b - - 0 18
Gull has just doubled its rooks on the c file and sees half a pawn white edge, SF only 20cps.
2r1r1k1/1b3np1/2n1p1qp/ppRp4/3P1P2/PP2QNPB/1B2P2K/2R5 b - - 0 29
SF now sees white is winning, but it is already too late. The battle is decided mainly by the white doubled rooks that support tactically a wide range of lines that would otherwise be impossible. That is why it is always good to have doubled rooks, even if you see nothing concrete, something is bound to exist.
SF would do well to think over why doubled rooks code does not work in it. This is a major term and SF loses games because of its lack, but probably some redundancies again, or imperfect definitions, or maybe even search, influences such an outcome. People that know SF code should know better.
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
1r6/4r1k1/1pbp2p1/q1p4p/P1Pp1Pn1/3P1BP1/1R1N3P/RQ4K1 b - - 0 61
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
4r3/4r1k1/1p1p2p1/q1p4p/P1Pp1Pn1/3P1NP1/1R5P/RQ4K1 w - - 0 63
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
q3r3/6k1/1R1p2p1/2p4p/P1Pp1PnN/3P2P1/4r2P/RQ4K1 w - - 0 65
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
2r2rk1/pb2q1pp/1pn1p3/3p4/3Pnp2/PP1Q1NP1/1BR1PPBP/2R3K1 b - - 0 18
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
2r1r1k1/1b3np1/2n1p1qp/ppRp4/3P1P2/PP2QNPB/1B2P2K/2R5 b - - 0 29
SF would do well to think over why doubled rooks code does not work in it. This is a major term and SF loses games because of its lack, but probably some redundancies again, or imperfect definitions, or maybe even search, influences such an outcome. People that know SF code should know better.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Re: Hacky rules 2
2 other games where a rook is much stronger than an enemy minor plus pawns, but SF would not think so.
6k1/1p5p/2p3p1/p1n5/2Pn4/4R2B/PP5K/8 w - - 0 42
It is interesting that both engines do not evaluate quite correctly this position, seeing only some 20cps white advantage. Gull is a bit better in its eval, but just by cps.
5k2/7R/8/ppn3p1/3n2Bp/1P6/P6K/8 b - - 0 48
Now SF is faster to recognize white is much better, but that does not help a lot, as the position is already lost. A good eval knowledge moves earlier, however, before the vital exchanges were made, could have saved SF much trouble and the point.
4k3/8/8/pp4p1/4n1Bp/1P1R4/P3n2K/8 b - - 0 51
A rook seems to be so much stronger than N + 2 pawns.
1k6/1p3p2/p2b2p1/5b2/8/1P1NR3/P1P5/1K6 w - - 0 40
It is rook vs minor + pawn. Gull sees half a pawn white edge, SF only 25cps.
4R3/1k3p2/p3b1p1/1p4b1/1N6/1P6/P1P1K3/8 w - - 0 54
Same imbalance. Moves later search already helps SF realise white is winning.
8/8/1k3ppb/p1N2b2/2K1R3/1P6/P7/8 w - - 0 67
Once again, rook proves to be much stronger than minor + pawn.
Well, you see SF imbalance eval, what concerns rook-minor imbalances, is really wrong. Games where SF evaluates incorrectly this imbalance are really rife. I posted just a few, but there are much more. Something really should be done to fix this problem.
One of Reuven's patches started well, maybe try to improve on it?
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
6k1/1p5p/2p3p1/p1n5/2Pn4/4R2B/PP5K/8 w - - 0 42
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
5k2/7R/8/ppn3p1/3n2Bp/1P6/P6K/8 b - - 0 48
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
4k3/8/8/pp4p1/4n1Bp/1P1R4/P3n2K/8 b - - 0 51
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
1k6/1p3p2/p2b2p1/5b2/8/1P1NR3/P1P5/1K6 w - - 0 40
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
4R3/1k3p2/p3b1p1/1p4b1/1N6/1P6/P1P1K3/8 w - - 0 54
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
8/8/1k3ppb/p1N2b2/2K1R3/1P6/P7/8 w - - 0 67
Well, you see SF imbalance eval, what concerns rook-minor imbalances, is really wrong. Games where SF evaluates incorrectly this imbalance are really rife. I posted just a few, but there are much more. Something really should be done to fix this problem.
One of Reuven's patches started well, maybe try to improve on it?
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Gullible 2
2 more games where SF severely underestimates blocked pawn on the 6th rank.
r6r/3qnk2/p3npp1/1pp1p2p/P3P2N/1PpP1NP1/2P1QP1P/R4RK1 w - - 0 30
Gull sees 60cps black edge, SF only 20cps. The c3 blocked pawn on the 6th rank does not bother SF too much.
3r3r/3q1k2/p4p2/2p1p1pp/Pp1nP2P/1PpPN1P1/2P2PQK/R4R2 w - - 0 37
Even now, SF sees only 40cps black edge, Gull 80cps. The c3 pawn is tremendous.
6r1/5k2/p3qr2/2p1pN1p/Pp1nP1p1/1PpP1PP1/2P2Q1K/4R1R1 w - - 0 44
Only now SF realises it is dead lost. The c3 blocked pawn is instrumental in claiming the win by the sheer pressure it exercises.
4r3/p1pk1p2/3p2p1/7r/4Pn2/1PB2P1p/PKP4P/3R1R2 b - - 0 31
Both engines see themselves in advantage. Who is right? I think Gull. The h3 black pawn is tremendous, and besides h2 is very vulnerable.
4r3/2pk4/p2p2p1/4rp2/4P3/1P1R1P1p/PKP3nP/2B2R2 w - - 0 35
SF still thinks it is equal, Gull sees half a pawn black advantage. The h3 blocked pawn is a permanent threat.
8/2pk4/p2p1P2/5r2/6p1/1PK2P1p/P1PB2nP/5R2 w - - 0 42
Only now SF realises it is losing.
Well, what to say, too many lost SF games on one and the same theme. I really already lost count of them. It should not be very difficult to fix this problem, but who knows when will this happen.
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
r6r/3qnk2/p3npp1/1pp1p2p/P3P2N/1PpP1NP1/2P1QP1P/R4RK1 w - - 0 30
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
3r3r/3q1k2/p4p2/2p1p1pp/Pp1nP2P/1PpPN1P1/2P2PQK/R4R2 w - - 0 37
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
6r1/5k2/p3qr2/2p1pN1p/Pp1nP1p1/1PpP1PP1/2P2Q1K/4R1R1 w - - 0 44
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
4r3/p1pk1p2/3p2p1/7r/4Pn2/1PB2P1p/PKP4P/3R1R2 b - - 0 31
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
4r3/2pk4/p2p2p1/4rp2/4P3/1P1R1P1p/PKP3nP/2B2R2 w - - 0 35
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
8/2pk4/p2p1P2/5r2/6p1/1PK2P1p/P1PB2nP/5R2 w - - 0 42
Well, what to say, too many lost SF games on one and the same theme. I really already lost count of them. It should not be very difficult to fix this problem, but who knows when will this happen.
-
carldaman
- Posts: 2283
- Joined: Sat Jun 02, 2012 2:13 am
Re: Huge simplification
Lyudmil Tsvetkov wrote:Perfect, that is precisely a KID-type closed position.mcostalba wrote:Following Lyudmil indications I have refined the detector requiring that pawns are on a diagonal chain. Now I don't detect anymore positions as the previous one posted, but only these kind of positions:mcostalba wrote: Ok I have pushed a branch called closed_positions:
https://github.com/mcostalba/Stockfish/ ... d_position
[d]rn1qk2r/pp3ppp/2p2n2/2Pp1b2/3Pp3/P1P1P3/3B1PPP/R2QKBNR w KQkq - 1 9
Taking into account that an f2 backward pawn is also present, this is precisely the type of position for which the white king should receive some nice penalty, as the general pattern favours black very much, especially what concerns possible king attacks.
I see SF starts understanding the KID now. Carl will be very happy.
I'm very happy even more so seeing that Marco is working with you on this very critical matter.
I believe he won't regret it 
Now, I've known that since about 5 months ago SF can play a mean KID (usually from the attacking/Black side) and some of the tweaks clearly found their way into how it plays these positions. It is mind boggling how every move has a purpose (see game below) and the execution is incredibly accurate, while Houdini or other top engines have no idea what's coming their way until it's too late.
The problem as I see it, is that SF's eval does not yet allow it to seek out these KID positions in the first place. Once faced with the well-known position after 12 moves in the game below, it outplays everybody in spectacular fashion.
The next step should be to bring its eval in line so that it fully recognizes the attacking potential of such positions and doesn't shy away from them. Rating such positions as much better for White is incorrect and counterproductive, since SF has tremendous fighting chances on the Black side of the Classical KID.
The game is _very typical_ of how SF plays the KID against top engines. I have numerous other examples.
40m/20m acc., 512 MB hash, 2 cores, 5-man syzygy, ponder off
| 8 | |||||||||
| 7 | |||||||||
| 6 | |||||||||
| 5 | |||||||||
| 4 | |||||||||
| 3 | |||||||||
| 2 | |||||||||
| 1 | |||||||||
| a | b | c | d | e | f | g | h |
r1b3k1/8/p5q1/3P4/P1P1P3/3BQPn1/R4KP1/7r b - - 4 42
Surreal, isn't it?
Cheers,
CL
-
carldaman
- Posts: 2283
- Joined: Sat Jun 02, 2012 2:13 am
Re: Huge simplification
Hi Joerg and Lyudmil,Lyudmil Tsvetkov wrote:Hi Joerg.Joerg Oster wrote:Black's pawn chain is pointing towards the kingside. White's chain towards the queenside. If Black castles short, there is no impact from white's chain. Contrary to black's chain, if White castles short.mcostalba wrote:Why it favors black and not white? What if position is flipped? It would favor white instead?Lyudmil Tsvetkov wrote: this is precisely the type of position for which the white king should receive some nice penalty, as the general pattern favours black very much, especially what concerns possible king attacks.
Once detected the closed position how can SF understand if it favors white or black?
So another idea might be to give a penalty for White castling short in these KID like positions. Or, if White already castled short, give a penalty to king safety.
Since this is partially redundant with pawns attacking kingRing, maybe that penalty should be increased in such positions.
Of course, that's all only valid if I correctly understood Lyudmil's idea.
As usual, you understood my idea better than myself.
If you allow me, I think the best thing with such structures would be to give penalty for the king in the general case when it is on the side where the chain is pointed at, i.e. for king on e1,f1,g1,h1, regardless of whether it has castled or not.
I do not know about the penalty, maybe 20cps for a start would be OK, but who knows. What were Joona's values?
You could also try a progressively larger penalty that increases from e1 to f1 to g1/h1. Maybe 10 cp for e1, 15 cp for f1 and 20 cp for g1/h1.
Regards,
CL