Stockfish endgame evaluation problem

Discussion of chess software programming and technical issues.

Moderator: Ras

jwes
Posts: 778
Joined: Sat Jul 01, 2006 7:11 am

Stockfish endgame evaluation problem

Post by jwes »

Stockfish evaluates this position as +8.20 for white. This seems like a very optimistic evaluation. It also seems that there should be some way search could see that it is not won. There are less than 20,000 relevant positions (bk on f8 or e7, white pieces anywhere, or without wp at f7 and bk on f8, g8 or h8, white pieces anywhere) and there should be some way the program can see that it cannot force any other positions.
[d]5k2/5Pp1/6P1/8/4K3/8/2B5/8 b - - 0 36
kinderchocolate
Posts: 454
Joined: Mon Nov 01, 2010 6:55 am
Full name: Ted Wong

Re: Stockfish endgame evaluation problem

Post by kinderchocolate »

To make your conclusion, the engine might need over 20 depth and quite impractical.
Michel
Posts: 2292
Joined: Mon Sep 29, 2008 1:50 am

Re: Stockfish endgame evaluation problem

Post by Michel »

To make your conclusion, the engine might need over 20 depth and quite impractical.
No he is suggesting on the fly bitboard generation. This does not need any depth and is entirely practical. This is how the Freezer utility works (http://www.freezerchess.com/).

Unfortunately it is hard for the engine (not assisted by a human) to know when it should use this method and what question should be solved by it (in this case: "Can the king be driven away from f8/e7?").
zamar
Posts: 613
Joined: Sun Jan 18, 2009 7:03 am

Re: Stockfish endgame evaluation problem

Post by zamar »

jwes wrote:Stockfish evaluates this position as +8.20 for white. This seems like a very optimistic evaluation.
White has an extra piece and a passed pawn on 7th rank. With very few exceptions this is always winning, so I see nothing wrong with the evaluation.
It also seems that there should be some way search could see that it is not won. There are less than 20,000 relevant positions (bk on f8 or e7, white pieces anywhere, or without wp at f7 and bk on f8, g8 or h8, white pieces anywhere) and there should be some way the program can see that it cannot force any other positions.
[d]5k2/5Pp1/6P1/8/4K3/8/2B5/8 b - - 0 36
For human it's easy to detect these kind of fortress positions. But for computers... nope.
Joona Kiiski
lech
Posts: 1169
Joined: Sun Feb 14, 2010 10:02 pm

Re: Stockfish endgame evaluation problem

Post by lech »

I am working at this. I believe that it is possible to solve it. :lol:
example:
[d] 7k/6rp/6pN/6P1/8/2B5/6r1/K7 b - - 0 1

Code: Select all

Sfx:
   1	00:00	          85	0	+6,10	Rg2g1+
   2	00:00	         142	0	+4,80	Rg2g1+ Ka1b2 Rg1f1 Bc3xg7+ Kh8xg7
   2	00:00	         165	0	+5,85	Rg2f2 Bc3xg7+ Kh8xg7
   3	00:00	         214	0	+5,17	Rg2f2 Bc3xg7+ Kh8xg7 Nh6g4
   3	00:00	         281	0	+5,65	Rg2c2 Nh6f7+ Kh8g8 Bc3xg7 Kg8xg7
   4	00:00	         518	0	+5,57	Rg2c2 Bc3xg7+ Kh8xg7 Ka1b1 Rc2d2
   5	00:00	       1.107	0	+5,29	Rg2c2 Nh6f7+ Kh8g8 Bc3xg7 Kg8xg7 Nf7e5
   6	00:00	       1.698	113.200	+5,41	Rg2c2 Nh6f7+ Kh8g8 Bc3xg7 Kg8xg7 Nf7e5 Rc2e2
   7	00:00	       5.593	180.419	+5,49	Rg2c2 Nh6f7+ Kh8g8 Bc3xg7 Kg8xg7 Nf7d6 Rc2g2 Ka1b1 Rg2xg5 Kb1c2
   8-	00:00	      10.410	221.489	+5,33	Rg2c2 Bc3d4 Rc2c1+ Ka1b2
   8	00:00	      12.293	157.602	+5,33	Rg2c2 Bc3d4 Rc2e2 Nh6f7+ Kh8g8 Bd4xg7 Kg8xg7 Nf7d6 Re2c2
   9+	00:00	      13.948	178.820	+5,53	Rg2c2 Bc3d4 Rc2e2 Nh6f7+ Kh8g8 Bd4xg7 Kg8xg7 Nf7d6 Re2g2 Ka1b1 Rg2xg5 Kb1c2 h7h5
   9	00:00	      19.394	208.537	+5,33	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2c2 Ka1b1 Rc2e2 Nh6f7+ Kh8g8 Bd4xg7 Kg8xg7 Nf7d6 Re2e5 Kb1c2 Re5xg5
  10-	00:00	      21.745	199.495	+5,25	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2c2 Ka1b1 Rc2e2 Nh6f7+ Kh8g8 Bd4xg7 Kg8xg7 Nf7d6 Re2e5 Kb1c2 Re5xg5
  10-	00:00	      28.774	230.192	+5,17	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2c2 Ka1b1 Rc2e2 Kb1a1 Re2e1+ Ka1b2 Re1e2+ Kb2a1
  10	00:00	      32.829	234.492	+5,33	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2c2 Ka1b1 Rc2e2 Nh6f7+ Kh8g8 Bd4xg7 Kg8xg7 Nf7d6 Re2e5 Kb1c2 Re5xg5
  11-	00:00	      33.705	240.750	+5,17	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2c2 Ka1b1 Rc2e2 Kb1a1 Re2e7 Ka1b2
  11	00:00	      37.204	238.487	+5,33	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2c2 Ka1b1 Rc2e2 Nh6f7+ Kh8g8 Bd4xg7 Kg8xg7 Nf7d8 h7h5 g5xh6/ep+ Kg7xh6
  12-	00:00	      39.330	228.662	+5,25	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2c2 Ka1b1 Rc2e2 Kb1a1 Re2d2
  12	00:00	      44.259	236.679	+5,33	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2c2 Ka1b1 Rc2e2 Nh6f7+ Kh8g8 Bd4xg7 Kg8xg7 Nf7d8 h7h5 g5xh6/ep+ Kg7xh6
  13-	00:00	      46.799	250.262	+5,25	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2c2 Ka1b1 Rc2e2 Bd4c3 Re2f2 Bc3e5 Rf2d2 Nh6f7+ Kh8g8 Be5xg7 Kg8xg7 Nf7e5
  13-	00:00	      51.624	254.305	+5,17	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2c2 Ka1b1 Rc2e2 Kb1c1 Re2g2 Kc1b1
  13	00:00	      59.627	254.816	+5,13	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2c2 Ka1b1 Rc2e2 Nh6f7+ Kh8g8 Bd4xg7 Kg8xg7 Nf7d6 Re2e5 Kb1c2 Re5c5+ Kc2b3 Rc5xg5
  14+	00:00	      72.098	272.067	+5,33	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2e2 Ka1b1 Re2e4 Bd4c3 Re4f4 Kb1b2 Rf4f1 Bc3e5 Rf1f2+ Kb2b1 Rf2d2 Nh6f7+ Kh8g8 Be5xg7 Kg8xg7 Nf7e5
  14	00:00	      75.273	284.049	+5,33	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2e2 Ka1b1 Re2e4 Bd4c3 Re4f4 Kb1b2 Rf4f1 Bc3e5 Rf1f2+ Kb2c3 Rf2g2 Nh6f7+
  15	00:00	     100.144	305.317	+5,13	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2h2 Ka1b1 Rh2e2 Kb1a1 Re2d2
  16	00:00	     163.956	375.185	+5,21	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2h2 Ka1b1 Rh2e2 Bd4f6 Re2g2 Bf6d4 Rg2h2 Kb1c1 Rh2e2 Kc1d1 Re2g2 Nh6f7+ Kh8g8 Bd4xg7 Kg8xg7
  17-	00:00	     196.152	392.304	+5,01	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2h2 Ka1b1 Rh2e2 Bd4f6 Re2g2 Bf6d4 Rg2h2 Kb1c1 Rh2e2 Kc1d1 Re2g2 Nh6f7+ Kh8g8 Bd4xg7 Kg8xg7
  17	00:00	     255.623	419.742	+4,96	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Bc3d4 Rg2h2 Ka1b1 Rh2e2 Bd4f6 Re2f2 Kb1c1 Rf2g2 Bf6d4 Rg2a2 Kc1d1 Ra2a5 Nh6f7+
  18-	00:00	     356.885	447.785	+4,68	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Ka1b1 Rg2f2 Kb1c1 Rf2a2 Bc3d4 Ra2a5 Bd4f6 Ra5b5 Kc1c2 Rb5d5 Kc2b3 Rd5d1 Nh6f7+
  18	00:01	     728.232	496.070	+4,68	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Ka1b1 Rg2f2 Kb1c1 Rf2a2 Bc3d4 Ra2e2 Bd4c3 Re2e7 Bc3f6 Re7e6 Nh6f7+ Kh8g8 Bf6xg7 Kg8xg7
  19	00:01	     877.609	510.831	+4,72	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Ka1b1 Rg2f2 Bc3e5 Rf2e2 Be5f6 Re2f2 Bf6e5
  20	00:01	   1.012.299	518.330	+4,68	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Ka1b1 Rg2f2 Bc3e5 Rf2e2 Be5f6 Re2f2 Bf6e5
  21	00:02	   1.161.818	531.238	+4,68	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Ka1b1 Rg2f2 Bc3e5 Rf2e2 Be5f6 Re2e4 Kb1c2 Re4e6 Kc2d3 Re6d6+ Kd3e4
  22	00:02	   1.422.208	548.479	+4,64	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Ka1b1 Rg2f2 Bc3e5 Rf2e2 Be5f6 Re2e6 Bf6d4 Re6e4 Bd4f6 Re4e6
  23-	00:02	   1.660.891	562.441	+4,56	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2g2 Ka1b1 Rg2f2 Bc3e5 Rf2e2 Be5f6 Re2e6 Bf6d4 Re6e4 Bd4f6 Re4e6
  23-	00:03	   1.960.753	578.393	+4,48	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2d3 Bc3f6 Rd3f3 Bf6e5 Rf3e3 Be5f6 Re3e6 Bf6d4 Re6e4 Bd4f6 Re4f4 Bf6c3 Rf4c4 Bc3f6 Rc4c6 Bf6d4 Rc6c4 Bd4f6
  23	00:03	   2.154.587	584.374	+4,48	Rg2c2 Bc3d4 Rc2d2 Bd4c3 Rd2d3 Bc3f6 Rd3f3 Bf6d4 Rf3f4 Bd4c3 Rf4c4 Bc3f6 Rc4c6 Bf6d4 Rc6e6 Ka1b1 Re6e1+ Kb1b2 Re1f1 Kb2c3 Rf1c1+ Kc3d3 Rc1d1+ Kd3c3 Rd1xd4 Kc3xd4 Rg7a7 Kd4e4 Kh8g7 Ke4f4 Ra7a5 Kf4g4 Kg7f8 Kg4f4 Kf8e7 Nh6g8+ Ke7f7 Ng8h6+ Kf7g7
  24-	00:04	   2.647.711	605.191	+4,28	Rg2c2 Bc3d4 Rc2d2 Bd4e5 Rd2e2 Be5d4 Re2e4 Bd4f6 Re4b4 Ka1a2 Rb4b6 Bf6d4 Rb6b7 Bd4f6 Rb7b5 Bf6d4 Rb5b7
  24	00:05	   3.186.489	614.322	+4,29	Rg2c2 Bc3d4 Rc2d2 Bd4e5 Rd2e2 Be5d4 Re2e4 Bd4f6 Re4b4 Ka1a2 Rb4b6 Bf6d4 Rb6b7 Bd4f6 Rb7b5 Bf6c3 Rb5c5 Bc3f6 Rc5c6 Bf6d4 Rc6c4 Bd4f6 Rc4b4 Bf6e5 Rb4e4 Be5f6 Re4e2+ Ka2b1 Re2g2 Bf6c3 Rg2h2 Bc3e5 Rh2h5 Kb1c2
  25-	00:05	   3.484.494	624.685	+4,01	Rg2c2 Bc3d4 Rc2d2 Bd4e5 Rd2e2 Be5c3 Re2g2 Bc3d4 Rg2h2 Ka1b1 Rh2e2 Bd4f6 Re2e6 Bf6c3 Re6e3 Bc3f6 Re3e6
  25-	00:06	   3.843.354	627.486	+3,73	Rg2c2 Bc3d4 Rc2d2 Bd4e5 Rd2e2 Be5c3 Re2g2 Bc3d4 Rg2h2 Ka1b1 Rh2e2 Bd4f6 Re2e6 Bf6d4 Re6e2
  25	00:08	   5.334.317	628.750	+3,95	Rg2c2 Bc3d4 Rc2d2 Bd4e5 Rd2e2 Be5c3 Re2e4 Ka1b1 Re4e6 Kb1b2 Re6b6+ Kb2a2 Rb6c6 Bc3d4 Rc6c4 Bd4e5 Rc4e4 Be5c3 Re4h4 Ka2b2 Rh4a4 Bc3f6 Ra4c4 Bf6e5 Rc4c7 Kb2b1 Rc7c4 Kb1b2
  26	00:09	   5.807.182	638.643	+3,91	Rg2c2 Bc3d4 Rc2d2 Bd4e5 Rd2e2 Be5c3 Re2e4 Ka1b1 Re4e6 Kb1b2 Re6b6+ Kb2a2 Rb6c6 Bc3e5 Rc6c4 Ka2b2 Rc4c5 Nh6f7+ Kh8g8 Nf7h6+ Kg8f8 Be5d6+
  27	00:10	   6.800.022	652.468	+3,91	Rg2c2 Bc3d4 Rc2d2 Bd4e5 Rd2e2 Be5c3 Re2e4 Ka1b1 Re4f4 Kb1b2 Rf4c4 Bc3f6 Rc4c6 Bf6e5 Rc6c5 Be5f6 Rc5c6
  28-	00:10	   7.049.869	655.801	+3,83	Rg2c2 Bc3d4 Rc2d2 Bd4f6 Rd2f2 Bf6d4 Rf2f3 Bd4e5 Rf3e3 Be5d4 Re3e4 Bd4f6 Re4b4 Bf6e5 Rb4b7 Ka1a2 Rb7a7+ Ka2b2 Ra7e7 Be5d4 Re7b7+ Kb2c3 Rb7b5 Bd4f6 Rb5b6 Bf6d4 Rb6b5
  28-	00:11	   7.339.647	660.693	+3,75	Rg2c2 Bc3d4 Rc2d2 Bd4f6 Rd2f2 Bf6d4 Rf2f3 Bd4e5 Rf3e3 Be5d4 Re3d3 Bd4f6 Rd3f3 Bf6d4 Rf3f4 Bd4e5 Rf4f5 Be5c3 Rf5c5 Bc3d4 Rc5c4 Bd4e5 Rc4h4 Ka1b2 Rh4e4 Be5f6 Re4f4 Bf6e5 Rf4f5 Be5d4 Rf5b5+ Kb2c2 Rb5xg5 Nh6f7+
  28-	00:12	   8.139.894	647.977	+3,59	Rg2c2 Bc3d4 Rc2c4 Bd4e5 Rc4e4 Be5f6 Re4h4 Ka1b2 Rh4f4 Bf6e5 Rf4e4 Be5f6 Re4e3 Kb2c2 Re3e6 Bf6d4 Re6e7 Kc2b2 Re7e2+ Kb2c3 Re2e1 Bd4f6 Re1f1 Bf6d4 Rf1f3+ Kc3c2 Rf3f5 Nh6xf5 g6xf5 Kc2d3 Kh8g8 Bd4xg7 Kg8xg7 Kd3e3 Kg7g8 Ke3d3 Kg8f8 Kd3d4
  28-	00:14	   9.035.025	641.100	+3,27	Rg2c2 Bc3d4 Rc2c4 Bd4e5 Rc4e4 Be5f6 Re4e1+ Ka1b2 Re1e2+ Kb2c3 Re2e6 Bf6d4 Re6e7 Bd4f6 Re7b7 Bf6d4 Rb7e7
  28-	00:16	  10.157.335	628.081	+2,62	Rg2c2 Bc3d4 Rc2c4 Bd4e5 Rc4e4 Be5f6 Re4e2 Bf6c3 Re2h2 Bc3e5 Rh2h3 Ka1b2 Rh3h4 Be5f6 Rh4h2+ Kb2c3 Rh2h3+ Kc3c2 Rh3f3 Bf6e5 Rf3f8 Kc2c3 Rf8c8+ Kc3b2 Rc8c5 Nh6f7+ Kh8g8 Nf7h6+ Kg8f8 Be5d6+
  28	02:37	  78.265.834	498.460	+3,83	Rg2xg5 Nh6f7+ Kh8g8 Nf7xg5 Rg7b7 Ka1a2 h7h6 Ng5e4 h6h5 Ne4f6+ Kg8f7 Nf6e4 Rb7a7+ Ka2b2 Kf7e6 Bc3d4 Ra7a5 Kb2c3 h5h4 Bd4b6 Ra5e5 Kc3d3 h4h3 Bb6g1 Re5a5 Kd3e3 Ra5a2 Ne4g5+ Ke6f5 Ng5f3 g6g5 Bg1h2 g5g4 Nf3d4+ Kf5g5 Bh2f4+ Kg5h5 Nd4e2 Kh5h4
lech
Posts: 1169
Joined: Sun Feb 14, 2010 10:02 pm

Re: Stockfish endgame evaluation problem

Post by lech »

2nd example (without a waste of strange)
(P IV 3GHz 2 thread.)

[d] 8/P7/4k3/8/5P2/4Bq2/5P2/5K2 b - - 0 1

Code: Select all

Sfx:
   1	00:00	         142	455	+5,41	Ke6d6
   2	00:00	         211	676	+5,17	Ke6d6 f4f5
   3	00:00	         466	1.493	+5,37	Ke6d6 f4f5 Qf3a8
   4	00:00	       1.581	5.067	+5,09	Ke6d6 Be3d4 Qf3e4 Bd4e5+ Kd6c5
   5	00:00	       3.345	10.198	+5,21	Ke6d6 Kf1e1 Kd6c7 f4f5 Qf3a8
   6-	00:00	       4.747	14.472	+4,92	Ke6d6 Kf1e1 Kd6c7 Ke1d2 Qf3a8 f4f5
   6	00:00	       7.465	19.906	+4,72	Ke6d6 Kf1e1 Kd6c7 Ke1d2 Qf3d5+ Kd2c3 Qd5a8
   7+	00:00	      12.582	32.261	+5,29	Ke6d6 Kf1e1 Kd6c7 Ke1d2 Kc7b7 Kd2d3 Kb7a8
   7	00:00	      13.126	33.656	+4,96	Ke6d6 Kf1e1 Kd6c7 Ke1d2 Kc7b7 Kd2d3 Kb7a8
   8	00:00	      25.101	57.439	+5,01	Ke6d6 f4f5 Kd6e5 Kf1e1 Ke5xf5 Ke1d2 Kf5e5 Kd2d3 Qf3e4+ Kd3c3
   9+	00:00	      31.978	70.591	+5,17	Ke6d6 f4f5 Kd6e5 Kf1e1 Ke5xf5 Ke1d2 Kf5e5 Kd2d3 Qf3e4+ Kd3c3 Ke5d6
   9	00:00	      40.307	71.720	+5,05	Ke6d6 Kf1e1 Kd6c7 Ke1d2 Kc7b7 Kd2d3 Qf3d5+ Be3d4 Kb7a8 Kd3e3 Qd5f5 f2f3 Qf5d5
  10-	00:00	      66.840	104.437	+4,96	Ke6d6 Be3b6 Qf3a8 Kf1e2 Kd6d5 f4f5
  10+	00:00	      83.066	123.610	+5,13	Ke6d6 Be3b6 Qf3e4 Bb6e3 Kd6c7 f4f5 Kc7b7 a7a8B+ Kb7xa8 f5f6 Qe4f3 Be3g5 Ka8b7
  10	00:00	      87.709	127.669	+5,13	Ke6d6 Kf1e1 Kd6c7 Ke1d2 Kc7b7 Kd2d3 Qf3d5+ Be3d4 Qd5f5+ Kd3e3 Kb7a8 f2f3 Qf5d5
  11+	00:00	     100.775	140.159	+5,25	Ke6d6 Kf1e1 Kd6c7 Ke1d2 Kc7b7 Kd2d3 Qf3d5+ Be3d4 Qd5f5+ Kd3e3 Kb7a8 f2f3 Qf5d5
  11-	00:00	     138.966	177.933	+5,01	Ke6d6 Be3b6 Qf3e4 Kf1g1 Qe4a8 Kg1f1 Kd6c6 Bb6e3 Kc6b7 Kf1e2
  11+	00:00	     153.313	175.214	+5,37	Ke6d7 f4f5 Kd7c7 f5f6 Kc7b7 a7a8N Kb7xa8 Be3g5 Ka8b7
  11	00:00	     174.098	179.667	+5,17	Ke6d7 Kf1e1 Kd7c7 Ke1d2 Qf3d5+ Kd2c3 Kc7b7 Kc3b4 Qd5e4+ Kb4b3 Kb7a8 Kb3c3 Qe4e7
  12+	00:01	     200.024	197.067	+5,25	Ke6d7 Kf1e1 Kd7c7 Ke1d2 Qf3d5+ Kd2c3 Kc7b7 Kc3b4 Qd5e4+ Kb4b3 Qe4e8 Kb3c4 Qe8c6+ Kc4b4 Kb7a8 Kb4a5 Qc6e4
  12-	00:01	     212.099	202.577	+5,09	Ke6d7 Kf1e1 Kd7c7 Ke1d2 Qf3d5+ Kd2c3 Kc7b7 Kc3b4 Qd5e4+ Kb4b3 Qe4e8 Kb3c4 Qe8c6+ Kc4b4 Kb7a8 Kb4a5 Qc6e4 Ka5a6 Qe4b7+ Ka6a5 Qb7c6 Ka5b4 Ka8b7 Be3c5 Qc6e4+ Kb4b5 Qe4xf4
  12	00:01	     258.182	229.495	+5,05	Ke6d7 Kf1e1 Kd7c7 Ke1d2 Qf3d5+ Kd2c3 Kc7b7 Kc3b4 Qd5e4+ Kb4b3 Qe4e6+ Kb3c3 Qe6a6 Be3d4 Qa6a5+ Kc3c4 Qa5c7+ Kc4d5 Qc7xf4
  13+	00:01	     292.350	231.106	+5,21	Ke6d7 Kf1e1 Kd7c7 Ke1d2 Qf3d5+ Kd2c3 Kc7b7 Kc3b4 Qd5e4+ Kb4b3 Qe4e6+ Kb3c3 Qe6a6 Be3d4 Qa6a5+ Kc3c4 Qa5c7+ Kc4d5 Qc7xf4
  13	00:01	     314.702	226.404	+5,17	Ke6d7 Kf1e1 Kd7c7 Ke1d2 Qf3d5+ Kd2c3 Kc7b7 Kc3b4 Qd5e4+ Kb4b3 Qe4e6+ Kb3c3 Qe6c6+ Kc3b4 Qc6c2 Be3c5 Qc2e4+ Kb4b5 Qe4xf4
  14+	00:01	     392.434	256.325	+5,37	Ke6d7 Kf1e1 Kd7c7 Ke1d2 Qf3d5+ Kd2c3 Kc7b7 Kc3b4 Qd5e4+ Kb4b3 Qe4c6 Kb3b4 Qc6c2 Be3c5 Qc2e4+ Kb4b5 Qe4e2+ Kb5a5 Qe2d2+ Ka5b5 Qd2xf4
  14	00:01	     440.125	265.775	+5,17	Ke6d7 Kf1e1 Kd7c7 Ke1d2 Qf3d5+ Kd2c3 Kc7b7 Kc3b4 Qd5d3 Kb4a4 Qd3c4+ Ka4a5 Qc4b3 Be3c5 Qb3d5 Ka5b5 Qd5c6+ Kb5b4 Qc6e4+ Kb4b5 Qe4xf4 Bc5e3 Qf4f5+ Be3c5 Qf5e4
  15+	00:01	     532.218	281.596	+5,25	Ke6d7 Kf1e1 Kd7c7 Ke1d2 Qf3d5+ Kd2c3 Kc7b7 Kc3b4 Qd5d3 Kb4a4 Qd3c4+ Ka4a5 Qc4b3 Be3c5 Qb3d5 Ka5b5 Qd5c6+ Kb5b4 Qc6e4+ Kb4b5 Qe4xf4 Bc5e3 Qf4f5+ Be3c5 Qf5e4
  15+	00:02	     647.738	304.817	+5,33	Ke6d7 Kf1e1 Qf3d5 Ke1e2 Kd7c7 f4f5 Kc7b7 a7a8N Kb7xa8 f5f6 Qd5f5 Be3d4 Qf5e4+ Bd4e3 Qe4h4
  15+	00:02	     812.132	341.950	+5,81	Ke6d7 Kf1e1 Qf3d5 Ke1e2 Kd7c7 f4f5 Kc7b7 a7a8N Kb7xa8 f5f6 Qd5f5 Be3d4 Qf5e4+ Bd4e3 Qe4h4
  15	00:03	   1.111.573	370.524	+6,02	Ke6d7 Kf1e1 Qf3d5 Ke1e2 Kd7c7 f4f5 Kc7b7 f5f6 Qd5b5+ Ke2e1 Qb5a5+ Ke1e2 Qa5a6+ Ke2d2 Qa6xf6 Kd2d3 Qf6f5+ Kd3c4 Qf5e4+ Kc4b5 Qe4d3+ Kb5b4 Qd3d5 Kb4a4 Kb7a8 Ka4b4 Ka8b7
  16	00:03	   1.363.194	391.272	+6,02	Ke6d7 Kf1e1 Qf3d5 Ke1e2 Kd7c7 f4f5 Kc7b7 f5f6 Qd5b5+ Ke2d2 Qb5b2+ Kd2d3 Qb2xf6 Kd3c4 Qf6f5 Kc4b4 Qf5e4+ Kb4b5 Qe4d3+ Kb5b4 Qd3d5 Kb4c3 Qd5e4 Be3d4 Kb7a8 Kc3c4 Ka8b7 Kc4c3
  17	00:03	   1.622.775	410.517	+6,02	Ke6d7 Kf1e1 Qf3d5 Ke1e2 Kd7c7 f4f5 Kc7b7 f5f6 Qd5b5+ Ke2d2 Qb5b2+ Kd2d3 Qb2xf6 Kd3c4 Qf6f5 Kc4b4 Qf5e4+ Kb4b5 Qe4d3+ Kb5b4 Qd3d5 Kb4c3 Qd5e4 Be3d4 Kb7a8 Kc3c4 Ka8b7 Kc4c3
  18-	00:04	   2.129.723	463.587	+5,93	Ke6d7 Kf1e1 Qf3d5 Ke1e2 Kd7c7 Be3d4 Qd5e4+ Ke2f1 Kc7b7 Bd4e3 Qe4h1+ Kf1e2 Qh1d5 Ke2f1 Qd5c4+ Kf1g2 Kb7a8 Kg2f3 Qc4c6+ Kf3g4
  18-	00:04	   2.242.556	472.117	+5,85	Ke6d7 Kf1e1 Qf3d5 Ke1e2 Kd7c7 Be3d4 Qd5e4+ Ke2f1 Kc7b7 Bd4e3 Qe4h1+ Kf1e2 Qh1d5 Ke2f1 Qd5c4+ Kf1g2 Kb7a8 Kg2f3 Qc4c6+ Kf3g4
  18-	00:05	   2.502.800	495.898	+5,37	Ke6d7 Kf1e1 Qf3d5 Ke1e2 Kd7c7 Be3d4 Qd5e4+ Ke2f1 Kc7b7 Bd4e3 Qe4h1+ Kf1e2 Qh1d5 Ke2f1 Qd5c4+ Kf1g2 Kb7a8 Kg2f3 Qc4c6+ Kf3g4
  18	00:05	   3.148.578	534.563	+5,24	Ke6d7 Kf1e1 Qf3d5 Ke1e2 Kd7c7 Be3d4 Qd5e4+ Ke2f1 Kc7b7 Bd4e3 Qe4h1+ Kf1e2 Qh1d5 Ke2f1 Qd5d1+ Kf1g2 Qd1g4+ Kg2f1 Kb7a8 Kf1e1 Qg4e6 Ke1f1 Qe6e4 Kf1e2 Qe4d5 Ke2f1 Qd5e4
  19	00:06	   3.720.162	566.925	+5,06	Ke6d7 Kf1e1 Qf3d5 Ke1e2 Kd7c7 Be3d4 Qd5e4+ Ke2f1 Kc7b7 Bd4e3 Qe4h1+ Kf1e2 Qh1d5 Ke2f1 Qd5d1+ Kf1g2 Qd1g4+ Kg2f1 Kb7a8 Kf1e1 Qg4e6 Ke1f1 Qe6e4 Kf1e2 Ka8b7 Ke2f1
  20	00:07	   4.497.705	600.976	+4,86	Ke6d7 Kf1e1 Qf3d5 Ke1e2 Kd7c7 Be3d4 Qd5e4+ Ke2f1 Kc7b7 Bd4e3 Qe4h1+ Kf1e2 Qh1d5 Ke2f1 Kb7a8 Kf1e2 Qd5e4 Ke2f1 Qe4f3 Kf1e1 Qf3c6 Ke1e2 Qc6d5 Ke2f1 Qd5f3
  21-	00:09	   6.093.376	637.249	+4,58	Ke6d7 Kf1e1 Qf3d5 Ke1e2 Kd7c7 Be3d4 Qd5e4+ Ke2f1 Kc7b7 Bd4e3 Qe4h1+ Kf1e2 Qh1d5 Ke2f1 Kb7a8 Kf1e2 Qd5e4 Ke2f1 Qe4f3 Kf1e1 Qf3c6 Ke1e2 Qc6d5 Ke2f1 Qd5f3
  21	00:12	   8.264.702	651.430	+5,01	Qf3d5 Kf1e1 Ke6f5 Ke1e2 Kf5g4 Be3d4 Kg4xf4 Ke2d3 Qd5b3+ Kd3d2 Qb3a3 Bd4e3+ Kf4e5
Jouni
Posts: 3651
Joined: Wed Mar 08, 2006 8:15 pm
Full name: Jouni Uski

Re: Stockfish endgame evaluation problem

Post by Jouni »

Thread title is misleading, because all engines I tested score from +5 to +9 in this position, even bigger than Stockfish:

Analysis by spark-0.4:

1...Ke7 2.Ke5 Kf8 3.Be4 Ke7 4.Bb1 Kf8 5.Bf5 Ke7 6.Bd3 Kf8 7.Kd5 Ke7 8.Bb1 Kf8 9.Be4 Ke7 10.Kc6 Kf8 11.Bh1 Ke7 12.Bd5 Kf8 13.Kb6 Ke7 14.Kc7 Kf8 15.Bg2 Ke7 16.Be4 Kf8 17.Bh1
+- (8.80) Depth: 33/42 00:00:05 45884kN

Jouni
mcostalba
Posts: 2684
Joined: Sat Jun 14, 2008 9:17 pm

Re: Stockfish endgame evaluation problem

Post by mcostalba »

The more I read threads like this the more I make up my mind that tweaking evaluation based on a given position is "the wrong thing to do" (tm) :D
lech
Posts: 1169
Joined: Sun Feb 14, 2010 10:02 pm

Re: Stockfish endgame evaluation problem

Post by lech »

Next interesting example (very hard for engines too):
[d] 7R/6pb/6k1/Pr4Pp/7P/6K1/8/8 b - - 0 1

Code: Select all

Sfx:
   1	00:00	          81	1.038	+4,00	Rb5xa5
   2	00:00	         135	1.730	+3,95	Rb5xa5 Kg3f3
   3	00:00	         284	3.641	+4,00	Rb5xa5 Kg3f3 Ra5a4
   4	00:00	       1.019	13.064	+4,08	Rb5xa5 Rh8f8 Ra5a3+ Kg3f2 Ra3a4
   5	00:00	       1.543	19.782	+4,12	Rb5xa5 Rh8f8 Ra5a3+ Rf8f3 Ra3xf3+ Kg3xf3 Bh7g8
   6	00:00	       1.974	21.000	+4,12	Rb5xa5 Rh8f8 Ra5a3+ Rf8f3 Ra3xf3+ Kg3xf3 Kg6f7 Kf3f4
   7+	00:00	       2.796	29.744	+4,20	Rb5xa5 Rh8f8 Ra5a3+ Rf8f3 Ra3xf3+ Kg3xf3 Kg6f7 Kf3f4 Bh7d3
   7+	00:00	       3.257	34.648	+4,28	Rb5xa5 Rh8f8 Ra5a3+ Rf8f3 Ra3xf3+ Kg3xf3 Kg6f7 Kf3f4 Bh7d3
   7	00:00	       5.653	45.224	+4,36	Rb5xa5 Rh8f8 Ra5a3+ Kg3f2 Ra3a4 Kf2g3 Ra4g4+ Kg3h3 Rg4a4
   8	00:00	      12.843	91.085	+4,40	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Rf8c8 Rb3b4 Kg2g3 Rb4g4+ Kg3h3 Rg4b4 Rc8c7 Bh7g8
   9-	00:00	      15.244	97.717	+4,24	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Rf8c8 Rb3b4 Rc8c6+ Kg6f5 Rc6c7
   9	00:00	      23.292	135.418	+4,36	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Rf8c8 Rb3b4 Kg2g3 Rb4g4+ Kg3h3 Rg4b4 Rc8c7 Bh7g8
  10-	00:00	      25.568	136.000	+4,25	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3c3 Kh2g2 Rc3b3
  10-	00:00	      28.250	150.265	+4,20	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3c3 Kh2g2 Rc3b3
  10	00:00	      31.616	168.170	+4,24	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3c3 Kh2g2 Rc3b3
  11	00:00	      38.718	190.729	+4,24	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3c3 Kh2g2 Rc3b3
  12	00:00	      56.083	224.332	+4,24	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3c3 Kh2g2 Rc3e3 Kg2h2 Re3e4 Kh2g3 Re4g4+ Kg3h3 Rg4b4 Kh3g3 Rb4b3+ Kg3h2
  13	00:00	      82.231	262.718	+4,24	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3b4 Kh2g3 Rb4a4 Rf8c8 Ra4g4+ Kg3h3 Rg4e4 Kh3g3 Kg6f7 Rc8c7+ Re4e7 Rc7xe7+ Kf7xe7 Kg3f4
  14	00:00	     134.658	319.094	+4,24	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3e3 Kh2g2 Re3c3 Rf8h8 Rc3c4 Kg2g3 Rc4g4+ Kg3h3 Rg4f4 Kh3g3 Rf4g4+
  15-	00:00	     158.616	338.200	+4,16	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3e3 Kh2g2 Re3c3 Rf8h8 Rc3c4 Kg2g3 Rc4g4+ Kg3h3 Rg4f4 Kh3g3 Rf4c4 Rh8b8 Rc4c3+ Kg3f2 Kg6f5
  15	00:00	     200.562	366.658	+4,16	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3e3 Kh2g2 Re3c3 Rf8h8 Rc3c4 Kg2g3 Rc4b4 Rh8f8 Rb4b7 Kg3g2 Rb7f7 Rf8h8 Rf7b7 Rh8f8
  16	00:00	     292.074	415.467	+4,15	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3e3 Kh2g2 Re3c3 Rf8h8 Rc3c4 Kg2g3 Rc4c3+ Kg3g2
  17+	00:01	     547.440	467.098	+4,23	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8c8 Rg4d4 Rc8c6+ Kg6f5 Rc6c7 Rd4d3+ Kh3h2 Kf5g4 Rc7xg7 Bh7e4
  17	00:01	     680.434	489.168	+4,11	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4a4 Rh8f8 Ra4a3+ Kg3g2 Ra3a2+ Kg2g3 Ra2a4
  18-	00:01	     717.769	493.991	+4,03	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4a4 Rh8f8 Ra4a3+ Kg3g2 Ra3a2+ Kg2g3 Ra2a4
  18-	00:01	     757.194	494.574	+3,94	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1h1 Rf8h8 Rh1a1 Kg3g2 Ra1a2+ Kg2g3 Ra2a3+ Kg3g2 Ra3a4 Kg2g3 Ra4a3+
  18	00:01	     815.363	501.761	+3,92	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1e5 Kg3g2 Re5d5 Kg2g3 Rd5d7 Kg3g2 Rd7b7 Kg2g3 Rb7f7 Rf8h8 Rf7d7 Kg3g2 Rd7d2+ Kg2g3 Rd2d3+ Kg3h2 Rd3b3 Kh2g2 Rb3a3 Rh8f8 Ra3a4 Kg2g3 Ra4a2 Rf8h8 Ra2a4 Rh8f8
  19	00:01	     882.105	504.060	+3,92	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1e5 Kg3g2 Re5e7 Kg2g3 Re7f7 Rf8h8 Rf7b7 Rh8f8 Rb7a7 Kg3g2 Ra7a2+ Kg2g3 Ra2a4
  20+	00:02	   1.036.875	514.322	+4,05	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1e5 Kg3g2 Re5e7 Kg2g3 Re7f7 Rf8h8 Rf7d7 Kg3g2 Rd7c7 Kg2g3 Rc7e7 Rh8f8
  20	00:02	   1.188.491	535.597	+3,88	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1g1+ Kg3h2 Rg1c1 Kh2g3 Rc1g1+
  21+	00:02	   1.270.450	538.554	+3,96	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1g1+ Kg3h3 Rg1c1 Kh3g2 Rc1c7 Kg2g3 Rc7c3+ Kg3g2 Rc3c4 Kg2g3 Rc4a4
  21+	00:02	   1.417.653	546.512	+4,05	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1g1+ Kg3h3 Rg1c1 Kh3g2 Rc1c4 Kg2g3 Rc4g4+ Kg3h3 Rg4b4 Kh3g3 Rb4c4 Kg3h3 Rc4b4
  21	00:02	   1.460.743	549.978	+4,04	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1g1+ Kg3h3 Rg1c1 Kh3g2 Rc1c4 Kg2h3 Rc4g4
  22	00:03	   1.943.893	568.057	+4,03	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1g1+ Kg3h3 Rg1c1 Kh3g2 Rc1c4 Kg2h3 Rc4b4 Rf8h8 Rb4g4 Rh8f8 Rg4d4 Rf8h8 Rd4g4
  23	00:04	   2.551.434	591.568	+4,02	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1g1+ Kg3h3 Rg1c1 Kh3g2 Rc1c4 Kg2g3 Rc4g4+ Kg3h3 Rg4b4 Kh3g3 Rb4c4 Rf8h8 Rc4g4+ Kg3h3
  24-	00:04	   2.744.046	597.310	+3,94	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Kg3f2 Re1e7 Kf2g3 Re7e3+ Kg3g2 Re3e2+ Kg2g3 Re2e3+
  24-	00:07	   4.304.292	595.008	+3,86	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4a4 Kh3g3 Ra4g4+ Kg3h3 Rg4d4 Rf8h8 Rd4b4
  24-	00:08	   4.937.562	593.956	+3,70	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4c4 Kh3g3 Rc4a4
  24	00:11	   6.562.497	578.499	+3,83	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4c4 Kh3g3 Rc4c2 Kg3h3 Rc2c4
  25	00:11	   7.055.985	589.521	+3,79	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4c4 Kh3g3 Rc4c3+ Kg3g2 Rc3e3 Kg2h2 Re3d3 Kh2g2 Rd3d2+ Kg2g3 Rd2a2 Kg3h3 Ra2d2 Kh3g3
  26+	00:13	   8.104.919	588.122	+3,95	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4c4 Kh3g3 Rc4c3+ Kg3g2 Rc3e3 Kg2h2 Re3d3 Kh2g2 Rd3d2+ Kg2g3 Rd2a2 Kg3h3 Ra2a4 Kh3g3 Ra4a5 Kg3g2 Ra5f5 Rf8h8 Rf5e5 Kg2g3 Re5e3+ Kg3h2 Re3e1 Kh2g3 Re1e2 Rh8c8 Kg6f7 Rc8c7+ Re2e7 Rc7xe7+ Kf7xe7 Kg3f4 Ke7e6 Kf4g3 Bh7f5 Kg3g2 Ke6f7 Kg2g3 Kf7e6
  26	00:14	   8.865.403	595.993	+3,86	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4c4 Kh3g3 Rc4a4
  27	00:16	   9.769.716	605.273	+3,85	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4c4 Kh3g3 Rc4g4+ Kg3h3 Rg4a4 Kh3g3 Ra4c4 Kg3h3 Rc4c2 Kh3g3 Rc2c4
  28-	00:21	  12.995.725	608.870	+3,77	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Rh8f8 Re4b4 Kh3g3 Rb4b5 Kg3g2 Rb5b2+ Kg2g3 Rb2c2 Kg3h3 Rc2c3+ Kh3h2 Rc3b3 Kh2g2 Rb3b2+
  28-	00:42	  23.240.018	552.098	+3,69	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4a2 Kh3g3 Ra2a3+ Kg3h2 Ra3a7 Kh2g3 Ra7a4
  28-	01:36	  51.713.991	538.597	+3,53	Rb5xa5 Rh8f8 Ra5b5 Kg3g2 Rb5b4 Kg2h3 Rb4g4 Rf8h8 Rg4b4 Rh8f8
  28-	01:44	  56.234.123	537.000	+3,21	Rb5xa5 Rh8f8 Ra5f5 Rf8h8 Rf5b5 Rh8f8 Rb5a5 Kg3g2 Ra5a4 Kg2g3 Ra4b4 Kg3h3 Rb4a4 Kh3g3
  28-	02:22	  74.297.519	521.843	+2,56	Rb5xa5 Rh8f8 Ra5f5 Rf8h8 Rf5b5 Rh8f8 Rb5a5 Kg3g2 Ra5e5 Kg2g3 Re5e1 Kg3g2 Re1c1 Kg2g3 Rc1h1 Rf8h8 Rh1f1 Kg3h3 Rf1f7 Kh3g3 Rf7f1
  28	06:48	 199.514.040	488.892	+3,75	Kg6f7 a5a6 Bh7e4 Kg3f4 Be4h1 Rh8c8 Rb5b4+ Kf4e3 Rb4e4+ Ke3f2 Re4xh4 Rc8c7+ Kf7g6 Rc7c1 Kg6xg5 a6a7 Bh1e4 Rc1c4 Rh4h2+ Kf2e3 Rh2a2 Rc4xe4 Ra2xa7 Re4e5+ Kg5g4 Re5e4+ Kg4f5 Re4h4 Ra7a3+ Ke3e2 Kf5g5 Rh4h1 h5h4 Rh1g1+ Kg5f6 Rg1f1+ Kf6e6 Rf1g1 Ra3a7 Ke2f3 Ke6f5 Rg1b1 Ra7a3+ Kf3g2 g7g5
It works by a function (5 lines) which modifies ss->eval.
For this reason it should'nt decrease ELO of Stockfish.
It needs many tests. :lol:
Uri Blass
Posts: 10889
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

Re: Stockfish endgame evaluation problem

Post by Uri Blass »

lech wrote:Next interesting example (very hard for engines too):
[d] 7R/6pb/6k1/Pr4Pp/7P/6K1/8/8 b - - 0 1

Code: Select all

Sfx:
   1	00:00	          81	1.038	+4,00	Rb5xa5
   2	00:00	         135	1.730	+3,95	Rb5xa5 Kg3f3
   3	00:00	         284	3.641	+4,00	Rb5xa5 Kg3f3 Ra5a4
   4	00:00	       1.019	13.064	+4,08	Rb5xa5 Rh8f8 Ra5a3+ Kg3f2 Ra3a4
   5	00:00	       1.543	19.782	+4,12	Rb5xa5 Rh8f8 Ra5a3+ Rf8f3 Ra3xf3+ Kg3xf3 Bh7g8
   6	00:00	       1.974	21.000	+4,12	Rb5xa5 Rh8f8 Ra5a3+ Rf8f3 Ra3xf3+ Kg3xf3 Kg6f7 Kf3f4
   7+	00:00	       2.796	29.744	+4,20	Rb5xa5 Rh8f8 Ra5a3+ Rf8f3 Ra3xf3+ Kg3xf3 Kg6f7 Kf3f4 Bh7d3
   7+	00:00	       3.257	34.648	+4,28	Rb5xa5 Rh8f8 Ra5a3+ Rf8f3 Ra3xf3+ Kg3xf3 Kg6f7 Kf3f4 Bh7d3
   7	00:00	       5.653	45.224	+4,36	Rb5xa5 Rh8f8 Ra5a3+ Kg3f2 Ra3a4 Kf2g3 Ra4g4+ Kg3h3 Rg4a4
   8	00:00	      12.843	91.085	+4,40	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Rf8c8 Rb3b4 Kg2g3 Rb4g4+ Kg3h3 Rg4b4 Rc8c7 Bh7g8
   9-	00:00	      15.244	97.717	+4,24	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Rf8c8 Rb3b4 Rc8c6+ Kg6f5 Rc6c7
   9	00:00	      23.292	135.418	+4,36	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Rf8c8 Rb3b4 Kg2g3 Rb4g4+ Kg3h3 Rg4b4 Rc8c7 Bh7g8
  10-	00:00	      25.568	136.000	+4,25	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3c3 Kh2g2 Rc3b3
  10-	00:00	      28.250	150.265	+4,20	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3c3 Kh2g2 Rc3b3
  10	00:00	      31.616	168.170	+4,24	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3c3 Kh2g2 Rc3b3
  11	00:00	      38.718	190.729	+4,24	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3c3 Kh2g2 Rc3b3
  12	00:00	      56.083	224.332	+4,24	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3c3 Kh2g2 Rc3e3 Kg2h2 Re3e4 Kh2g3 Re4g4+ Kg3h3 Rg4b4 Kh3g3 Rb4b3+ Kg3h2
  13	00:00	      82.231	262.718	+4,24	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3b4 Kh2g3 Rb4a4 Rf8c8 Ra4g4+ Kg3h3 Rg4e4 Kh3g3 Kg6f7 Rc8c7+ Re4e7 Rc7xe7+ Kf7xe7 Kg3f4
  14	00:00	     134.658	319.094	+4,24	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3e3 Kh2g2 Re3c3 Rf8h8 Rc3c4 Kg2g3 Rc4g4+ Kg3h3 Rg4f4 Kh3g3 Rf4g4+
  15-	00:00	     158.616	338.200	+4,16	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3e3 Kh2g2 Re3c3 Rf8h8 Rc3c4 Kg2g3 Rc4g4+ Kg3h3 Rg4f4 Kh3g3 Rf4c4 Rh8b8 Rc4c3+ Kg3f2 Kg6f5
  15	00:00	     200.562	366.658	+4,16	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3e3 Kh2g2 Re3c3 Rf8h8 Rc3c4 Kg2g3 Rc4b4 Rh8f8 Rb4b7 Kg3g2 Rb7f7 Rf8h8 Rf7b7 Rh8f8
  16	00:00	     292.074	415.467	+4,15	Rb5xa5 Rh8f8 Ra5a3+ Kg3g2 Ra3b3 Kg2h2 Rb3e3 Kh2g2 Re3c3 Rf8h8 Rc3c4 Kg2g3 Rc4c3+ Kg3g2
  17+	00:01	     547.440	467.098	+4,23	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8c8 Rg4d4 Rc8c6+ Kg6f5 Rc6c7 Rd4d3+ Kh3h2 Kf5g4 Rc7xg7 Bh7e4
  17	00:01	     680.434	489.168	+4,11	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4a4 Rh8f8 Ra4a3+ Kg3g2 Ra3a2+ Kg2g3 Ra2a4
  18-	00:01	     717.769	493.991	+4,03	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4a4 Rh8f8 Ra4a3+ Kg3g2 Ra3a2+ Kg2g3 Ra2a4
  18-	00:01	     757.194	494.574	+3,94	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1h1 Rf8h8 Rh1a1 Kg3g2 Ra1a2+ Kg2g3 Ra2a3+ Kg3g2 Ra3a4 Kg2g3 Ra4a3+
  18	00:01	     815.363	501.761	+3,92	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1e5 Kg3g2 Re5d5 Kg2g3 Rd5d7 Kg3g2 Rd7b7 Kg2g3 Rb7f7 Rf8h8 Rf7d7 Kg3g2 Rd7d2+ Kg2g3 Rd2d3+ Kg3h2 Rd3b3 Kh2g2 Rb3a3 Rh8f8 Ra3a4 Kg2g3 Ra4a2 Rf8h8 Ra2a4 Rh8f8
  19	00:01	     882.105	504.060	+3,92	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1e5 Kg3g2 Re5e7 Kg2g3 Re7f7 Rf8h8 Rf7b7 Rh8f8 Rb7a7 Kg3g2 Ra7a2+ Kg2g3 Ra2a4
  20+	00:02	   1.036.875	514.322	+4,05	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1e5 Kg3g2 Re5e7 Kg2g3 Re7f7 Rf8h8 Rf7d7 Kg3g2 Rd7c7 Kg2g3 Rc7e7 Rh8f8
  20	00:02	   1.188.491	535.597	+3,88	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1g1+ Kg3h2 Rg1c1 Kh2g3 Rc1g1+
  21+	00:02	   1.270.450	538.554	+3,96	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1g1+ Kg3h3 Rg1c1 Kh3g2 Rc1c7 Kg2g3 Rc7c3+ Kg3g2 Rc3c4 Kg2g3 Rc4a4
  21+	00:02	   1.417.653	546.512	+4,05	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1g1+ Kg3h3 Rg1c1 Kh3g2 Rc1c4 Kg2g3 Rc4g4+ Kg3h3 Rg4b4 Kh3g3 Rb4c4 Kg3h3 Rc4b4
  21	00:02	   1.460.743	549.978	+4,04	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1g1+ Kg3h3 Rg1c1 Kh3g2 Rc1c4 Kg2h3 Rc4g4
  22	00:03	   1.943.893	568.057	+4,03	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1g1+ Kg3h3 Rg1c1 Kh3g2 Rc1c4 Kg2h3 Rc4b4 Rf8h8 Rb4g4 Rh8f8 Rg4d4 Rf8h8 Rd4g4
  23	00:04	   2.551.434	591.568	+4,02	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Rh8f8 Re1g1+ Kg3h3 Rg1c1 Kh3g2 Rc1c4 Kg2g3 Rc4g4+ Kg3h3 Rg4b4 Kh3g3 Rb4c4 Rf8h8 Rc4g4+ Kg3h3
  24-	00:04	   2.744.046	597.310	+3,94	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Kh3g3 Re4e1 Kg3f2 Re1e7 Kf2g3 Re7e3+ Kg3g2 Re3e2+ Kg2g3 Re2e3+
  24-	00:07	   4.304.292	595.008	+3,86	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4a4 Kh3g3 Ra4g4+ Kg3h3 Rg4d4 Rf8h8 Rd4b4
  24-	00:08	   4.937.562	593.956	+3,70	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4c4 Kh3g3 Rc4a4
  24	00:11	   6.562.497	578.499	+3,83	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4c4 Kh3g3 Rc4c2 Kg3h3 Rc2c4
  25	00:11	   7.055.985	589.521	+3,79	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4c4 Kh3g3 Rc4c3+ Kg3g2 Rc3e3 Kg2h2 Re3d3 Kh2g2 Rd3d2+ Kg2g3 Rd2a2 Kg3h3 Ra2d2 Kh3g3
  26+	00:13	   8.104.919	588.122	+3,95	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4c4 Kh3g3 Rc4c3+ Kg3g2 Rc3e3 Kg2h2 Re3d3 Kh2g2 Rd3d2+ Kg2g3 Rd2a2 Kg3h3 Ra2a4 Kh3g3 Ra4a5 Kg3g2 Ra5f5 Rf8h8 Rf5e5 Kg2g3 Re5e3+ Kg3h2 Re3e1 Kh2g3 Re1e2 Rh8c8 Kg6f7 Rc8c7+ Re2e7 Rc7xe7+ Kf7xe7 Kg3f4 Ke7e6 Kf4g3 Bh7f5 Kg3g2 Ke6f7 Kg2g3 Kf7e6
  26	00:14	   8.865.403	595.993	+3,86	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4c4 Kh3g3 Rc4a4
  27	00:16	   9.769.716	605.273	+3,85	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4b4 Rh8f8 Rb4c4 Kh3g3 Rc4g4+ Kg3h3 Rg4a4 Kh3g3 Ra4c4 Kg3h3 Rc4c2 Kh3g3 Rc2c4
  28-	00:21	  12.995.725	608.870	+3,77	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4g4 Rf8h8 Rg4e4 Rh8f8 Re4b4 Kh3g3 Rb4b5 Kg3g2 Rb5b2+ Kg2g3 Rb2c2 Kg3h3 Rc2c3+ Kh3h2 Rc3b3 Kh2g2 Rb3b2+
  28-	00:42	  23.240.018	552.098	+3,69	Rb5xa5 Rh8f8 Ra5a4 Kg3h3 Ra4a2 Kh3g3 Ra2a3+ Kg3h2 Ra3a7 Kh2g3 Ra7a4
  28-	01:36	  51.713.991	538.597	+3,53	Rb5xa5 Rh8f8 Ra5b5 Kg3g2 Rb5b4 Kg2h3 Rb4g4 Rf8h8 Rg4b4 Rh8f8
  28-	01:44	  56.234.123	537.000	+3,21	Rb5xa5 Rh8f8 Ra5f5 Rf8h8 Rf5b5 Rh8f8 Rb5a5 Kg3g2 Ra5a4 Kg2g3 Ra4b4 Kg3h3 Rb4a4 Kh3g3
  28-	02:22	  74.297.519	521.843	+2,56	Rb5xa5 Rh8f8 Ra5f5 Rf8h8 Rf5b5 Rh8f8 Rb5a5 Kg3g2 Ra5e5 Kg2g3 Re5e1 Kg3g2 Re1c1 Kg2g3 Rc1h1 Rf8h8 Rh1f1 Kg3h3 Rf1f7 Kh3g3 Rf7f1
  28	06:48	 199.514.040	488.892	+3,75	Kg6f7 a5a6 Bh7e4 Kg3f4 Be4h1 Rh8c8 Rb5b4+ Kf4e3 Rb4e4+ Ke3f2 Re4xh4 Rc8c7+ Kf7g6 Rc7c1 Kg6xg5 a6a7 Bh1e4 Rc1c4 Rh4h2+ Kf2e3 Rh2a2 Rc4xe4 Ra2xa7 Re4e5+ Kg5g4 Re5e4+ Kg4f5 Re4h4 Ra7a3+ Ke3e2 Kf5g5 Rh4h1 h5h4 Rh1g1+ Kg5f6 Rg1f1+ Kf6e6 Rf1g1 Ra3a7 Ke2f3 Ke6f5 Rg1b1 Ra7a3+ Kf3g2 g7g5
It works by a function (5 lines) which modifies ss->eval.
For this reason it should'nt decrease ELO of Stockfish.
It needs many tests. :lol:
I believe that stockfish underevaluate mobility(of pieces with very small number of squares like the bishop at h7) so it overestimate black chances by not giving enough weight to the fact that the bishop h7 cannot move to more than one square but by comparing the evaluation at small depths
I see no modification of the evaluation because my stockfish1.9 get exactly the same evaluation as your analysis

I do not know what did you change but number of lines that you change in stockfish is evidence for nothing(and it is easy to decrease elo by changing one line of stockfish)


New game
7R/6pb/6k1/Pr4Pp/7P/6K1/8/8 b - - 0 1

Analysis by Stockfish 1.9 JA:

1...Rb5xa5
-+ (-4.00) Depth: 1 00:00:00
1...Rb5xa5 2.Kg3-f3
-+ (-3.95) Depth: 2 00:00:00
1...Rb5xa5 2.Kg3-f3 Ra5-a4
-+ (-4.00) Depth: 3 00:00:00
1...Rb5xa5 2.Rh8-f8 Ra5-a3+ 3.Kg3-f2 Ra3-a4
-+ (-4.08) Depth: 4 00:00:00
1...Rb5xa5 2.Rh8-f8 Ra5-a3+ 3.Rf8-f3 Ra3xf3+ 4.Kg3xf3 Bh7-g8
-+ (-4.12) Depth: 5 00:00:00