Why would it take most engines so long to see this is an easy win? Fruit takes minutes before it gets above +1. When I play it against Fruit at 40 moves/min it takes me to play more than 10 moves before Fruit's score gets significantly away from 0 (and drops to -12).
[d]8/6k1/1p6/6P1/p1p5/P1P2K1p/2P2P2/8 b - - 0 1
Stupid engines
Moderators: hgm, Rebel, chrisw
-
hgm
- Posts: 27825
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
-
Albert Silver
- Posts: 3019
- Joined: Wed Mar 08, 2006 9:57 pm
- Location: Rio de Janeiro, Brazil
Re: Stupid engines
Hash table bug? I tried the Big Five (Houdini, Komodo, Rybka, Critter, and Stockfish) and they all had large evals in seconds.hgm wrote:Why would it take most engines so long to see this is an easy win? Fruit takes minutes before it gets above +1. When I play it against Fruit at 40 moves/min it takes me to play more than 10 moves before Fruit's score gets significantly away from 0 (and drops to -12).
[d]8/6k1/1p6/6P1/p1p5/P1P2K1p/2P2P2/8 b - - 0 1
"Tactics are the bricks and sticks that make up a game, but positional play is the architectural blueprint."
-
hgm
- Posts: 27825
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
Re: Stupid engines
Well, Fruit 2.1 is not generally known for being buggy, but there surely seems something wrong. (Although I was told The King also has trouble with this position.) Fruit takes 19 min to reach 33ply here, and even then it is only at -1.40 for black:
The weird thing is that when I follow the PVs that score more that 1, it ends in positions like
[d]8/8/8/5k2/p1K5/P1P5/2P5/8 b
or
[d]8/8/8/8/p2K4/P1P5/2k5/8 w
Fruit does not do hash cuts in the PV, so the latter should always lead to the score-determining position. +1.4 for an extra doubled passer supported by King seems a gross underestimate (in KPPK this would be a guaranteed win), but for the second one, with the black King outside the passer's square, ('unstoppable passer') it is outright ridiculous.
Perhaps 16MB was is a bit on the low side, although in Pawn endings often very little hash is used. But that would explain the long times, but not the weird scoring.
Code: Select all
33 -1.40 1247.2M 19:19.99 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Ke6 Ke4 Kf6 Kd4 Kf5 Ke3 Ke6 f5+ Kf6 Kd4 Kf7 Kc5 Kf6 Kxb5 Kxf5 Kxc4
33 -1.40 1232.7M 19:02.48 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Ke6 Ke4 Kf6 Kd4 Kf5 Ke3 Ke6 f5+ Kf6 Kd4 Kf7 Kc5 Kf6 Kxb5 Kxf5 Kxc4
32 -1.40 1021.1M 14:44.25 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 f5 Ke5 f6 Kxf6 Kd4 Kf5 Kc5 Ke4 Kxb5 Kd5 Kb4 Kd6 Kxc4
31 -1.26 486.7M 6:58.04 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Kd6 Ke4 Ke6 f5+ Kf6 Kd5 Kxf5 Kc5 Ke4 Kxb5 Ke3 Kxc4 Kd2 Kd4 Kxc2
30 -0.92 173.7M 2:11.04 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 f5 Ke5 f6 Kxf6 Kd4 Ke7 Kc5 Ke6 Kxb5 Kd5 Kxa4
29 -0.92 159.9M 1:54.96 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 f5 Ke5 f6 Kxf6 Kd4 Ke7 Kc5 Ke6 Kxb5 Kd5 Kxa4
28 -0.70 35.3M 0:28.09 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Kd6 Ke4 Ke6 f5+ Kd6 Kd4 Kc6 Ke3 Kd5
27 -0.64 28.6M 0:22.25 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Kd6 Ke4 Ke6 f5+ Kf6 Kf4 Kg7 Ke3 Kf6
26 -0.71 26.1M 0:19.97 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Kd6 Ke4 Ke6 f5+ Kf6 Kf4 Ke7
25 -0.28 7.0M 0:05.18 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kg4 Ke2 Kf4 f3 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Ke6 Kf2 Kd6 Kg3 Ke6 Kg4
24 -0.32 6.1M 0:04.38 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kd5 Kf2 Kd6 Kg3 Ke6 Kf4 Kd5 Kf5 Kd6 Ke4 Ke6
23 -0.31 3.9M 0:02.88 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kd5 Kf2 Kd6 f4 Ke6 Ke2 Kf6 Kd2 Kf5 Ke3
22 -0.21 1.3M 0:01.17 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kd5 Kf2 Kd6 f4 Ke6 Ke2 Kf5 Kf3 Ke6
21 -0.21 1.2M 0:01.02 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Ke6
20 -0.19 596134 0:00.59 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kf5 Kf3
19 -0.19 484749 0:00.49 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke5 f4+ Kf5 Kf3
18 -0.08 243124 0:00.31 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Ke5 Ke3 Kd5
17 -0.08 186221 0:00.26 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Ke5 Ke3 Kd5
16 -0.08 147100 0:00.22 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Ke5 Ke3
15 -0.18 107571 0:00.19 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 f3 Ke5 Kg3 Kf5 f4 Ke4 Kg4 Kd5
14 -0.16 72543 0:00.15 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 f3 Ke5 Kg3 Kf5 f4 Ke4 Kg4
13 0.00 46988 0:00.10 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 f3 Ke5 Kf2 Kf4 Kg2
12 -0.08 32740 0:00.09 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 f3 Ke5 Kf2 Kd5 Ke3
11 +0.16 21116 0:00.08 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 f3 Ke5 Kg3 Kd5
10 +0.04 13981 0:00.06 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Ke5 Ke2
9 +0.28 8457 0:00.02 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 f3 Ke5
8 +0.16 5112 0:00.01 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 Ke4 Kg3
7 +0.28 2929 0:00.01 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 Ke4
6 +0.16 1595 0:00.00 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2
5 +0.28 755 0:00.00 Kg6 Kg3 h2 Kxh2 Kxg5
4 +0.28 391 0:00.00 Kg6 Kg3 h2 Kxh2 Kxg5
3 +0.28 250 0:00.00 Kg6 Kg3 h2 Kxh2 Kxg5
3 +0.24 206 0:00.00 h2 Kg2 Kg6 f4
2 +0.28 87 0:00.00 h2 Kg2 Kg6 Kxh2 Kxg5
2 +0.16 35 0:00.00 Kg6 Kg3 Kxg5 Kxh3
1 +0.26 14 0:00.00 Kg6
[d]8/8/8/5k2/p1K5/P1P5/2P5/8 b
or
[d]8/8/8/8/p2K4/P1P5/2k5/8 w
Fruit does not do hash cuts in the PV, so the latter should always lead to the score-determining position. +1.4 for an extra doubled passer supported by King seems a gross underestimate (in KPPK this would be a guaranteed win), but for the second one, with the black King outside the passer's square, ('unstoppable passer') it is outright ridiculous.
Perhaps 16MB was is a bit on the low side, although in Pawn endings often very little hash is used. But that would explain the long times, but not the weird scoring.
-
Daniel Shawul
- Posts: 4185
- Joined: Tue Mar 14, 2006 11:34 am
- Location: Ethiopia
Re: Stupid engines
I think that in pawn endgames one pawn up is significant advantage. Nebiyu/scorpio just add +100 to the eval score when that is the case. This is an idea I got from Ed schroeder and it works well in practise, and my engines do not have problem knowing black is in bad shape. What is probably making things hard for fruit is a) lack of that eval b) no hash cutoffs in PV c) the fact that black's blocked pawns are far more advanced than whites. Sjeng takes long to too so it is probably not only fruit related.
-
Michel
- Posts: 2272
- Joined: Mon Sep 29, 2008 1:50 am
Re: Stupid engines
Even GNU Chess 5.07.173b sees this is an easy win for white.
Code: Select all
vdbergh@pl:~/SRC/CHESS/GNUCHESS/gnuchess-git/src$ gnuchess -x
setboard 8/6k1/1p6/6P1/p1p5/P1P2K1p/2P2P2/8 b - - 0 1
analyze
1 -39 0 2 h2
2 9 0 19 h2!
2 9 0 30 h2 Kg2
2 73 0 34 Kg6!
2 -1 0 48 Kg6 Kg3
2 9 0 58 h2 Kg2
3 4 0 108 h2 Kg2 Kg6
4 4 0 213 h2 Kg2 Kg6 Kxh2
5 24 0 553 Kg6 Kg3 Kxg5 Kxh3 Kf4
6 -4 0 947 :-(
6 -4 0 1126 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2
7 25 0 1447 Kg6!
7 30 0 1629 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 Ke4
8 0 0 3410 :-(
8 20 1 4790 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 Ke4 Kg3
9 17 1 6975 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 Kg4 f3+ Kf5 Kg3 Ke5 Kg4
10 9 1 11265 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 Kg4 f3+ Kf5 Kg3 Ke5 f4+ Ke4 Kg4
11 -15 2 16738 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 Kg4 f3+ Kf5 Kg3 b5 f4 Ke4 Kg4 Kd5
12 -24 3 28569 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 Kd2 Kd5
13 0 3 38752 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 Kd2 Kf3 Ke1
14 0 4 53103 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 Kd2 Kf3 Ke1
15 0 5 62307 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 Kd2 Kf3 Ke1
16 -25 6 84264 :-(
16 0 9 110393 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 Kf1 Kf3
17 -25 11 133030 :-(
17 -31 13 159462 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Ke5 Ke3
18 -38 16 190937 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Ke5 Ke3 Kd5 f4
19 -64 20 253557 :-(
19 -89 25 339581 :-(
19 -114 29 399255 :-(
19 -44 35 482407 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kd5 Ke3 Ke5 f4+ Kd5 Kf3 Kc5
20 -72 37 532241 :-(
20 -97 39 558613 :-(
20 -122 42 610464 :-(
20 -147 45 666848 :-(
20 -172 50 746237 :-(
20 -97 55 845824 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kg5 Ke3 Kf6 Kd4 Kf5
21 -123 60 929379 :-(
21 -148 63 978384 :-(
21 -173 68 1064131 :-(
21 -124 74 1166207 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke6 Kd4 Kf5 Kc5 Ke5
22 -149 78 1246969 :-(
22 -174 82 1319346 :-(
22 -164 87 1401127 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke6 Ke4 Kf6 f4 Ke6 f5+ Kd6 Kd4 Ke7 Ke5
23 -189 92 1484458 :-(
23 -186 109 1765177 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke6 Ke4 Kf6 f4 Ke6 f5+ Kd6 f6
24 -211 120 1942904 :-(
24 -236 130 2134701 :-(
24 -211 142 2325389 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke6 Kd4 Kd6 f4 Kc6 Ke5 Kb6 f5 Kc6 f6 Kd7
25 -236 163 2618680 :-(
25 -261 184 2972888 :-(
25 -246 213 3395866 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke6 Kd4 Kd6 f4 Kc6 f5 Kd6 Ke4 Kc5 f6 Kd6 f7
26 -221 224 3553719 Kg6!
26 -209 234 3679275 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke6 Kd4 Kd6 f4 Kc6 Ke5 Kd7 f5 Ke7 f6+ Kf7
27 -234 245 3850334 :-(
27 -259 247 3892037 :-(
27 -186 294 4535802 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke6 Kd4 Kd6 f4 Kc6 f5 Kd6 Ke4 Kc5 f6 Kd6
28 -211 301 4652884 :-(
28 -236 309 4783388 :-(
28 -246 346 5279223 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke6 Kd4 Kd6 f4 Kc6 f5 Kd6 Ke4 Kc5 f6 Kd6 f7
29 -252 473 7050304 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke6 Kd4 Kd6 f4 Ke7 Ke5 Kd7 f5 Ke7 f6+ Kf7 Kf5 Kf8 Ke5
30 -277 511 7698898 :-(
30 -302 539 8170971 :-(
30 -281 600 9081445 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke5 f4+ Kd5 f5 Ke5 f6 Kxf6 Kd4 Ke6 Kc5 b4 cxb4 c3 Kd4 Kd6
31 -306 629 9606383 :-(
31 -331 692 10548441 :-(
31 -334 1021 15432888 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Ke6 Ke4 Kd6 Kd4 Ke6 Kc5 Kf6 Kxb5 Ke7 Kxc4 Kd6 f5
32 -359 1084 16434916 :-(
32 -384 1130 17225855 :-(
32 -369 1472 22092362 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Ke6 Kg4 Ke7 Kg5 Ke8 f5 b4 cxb4 Kd7 b5 Kd6 b6 c3
33 -394 1525 22991369 :-(
33 -419 1584 24012633 :-(
33 -444 1620 24662495 :-(
33 -469 1649 25195800 :-(
33 -494 1683 25816263 :-(
33 -519 1749 27011537 :-(
33 -544 1781 27627763 :-(
33 -569 1862 29077916 :-(
33 -565 1951 30521430 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke5 f4+ Kd5 f5 Ke5 f6 Kd6 Kf4 Ke6 Kg5 Kf7 Kf5 b4 cxb4 c3 Ke5 Kg6 b5
-
bob
- Posts: 20943
- Joined: Mon Feb 27, 2006 7:30 pm
- Location: Birmingham, AL
Re: Stupid engines
No idea why. Crafty spots this absolutely instantly:
15-> 0.02 0.45 1. ... Kg6 2. Kg3 Kxg5 3. Kxh3 Kf4
4. Kg2 Kg4 5. f3+ Kf4 6. Kf2 b5 7.
Ke2 Ke5 8. Ke3 Kf5 9. Kd4
19 0.06 0.91 1. ... Kg6 2. Kg3 Kxg5 3. Kxh3 Kf4
4. Kg2 b5 5. Kf1 Kf3 6. Ke1 Ke4 7.
Ke2 Kf4 8. f3 Kf5 9. Ke3 Ke5 10. f4+
Kf6 11. Ke4
23 0.17 1.09 1. ... Kg6 2. Kg3 Kxg5 3. Kxh3 Kf4
4. Kg2 b5 5. Kf1 Kf3 6. Ke1 Ke4 7.
Ke2 Kf4 8. f3 Kf5 9. Ke3 Ke5 10. f4+
Kd5 11. Kf3 Ke6 12. Ke4 Kf6 13. Kd5
25 0.34 1.72 1. ... Kg6 2. Kg3 Kxg5 3. Kxh3 Kf4
4. Kg2 b5 5. Kf1 Kf3 6. Ke1 Ke4 7.
Ke2 Kf4 8. f3 Kf5 9. Ke3 Ke5 10. f4+
Kd5 11. Kf3 Kd6 12. Ke4 Ke6 13. f5+
Kf6 14. Kf4 Kf7
30 3.58 2.52 1. ... Kg6 2. Kg3 Kxg5 3. Kxh3 Kf4
4. Kg2 b5 5. Kf1 Kf3 6. Ke1 Ke4 7.
Ke2 Kf4 8. f3 Kf5 9. Ke3 Ke5 10. f4+
Kd5 11. Kf3 Kd6 12. Ke4 Ke6 13. f5+
Kf6 14. Kf4 Kf7 15. Ke5 Ke7 16. f6+
Kd7 17. Kf5 Kd6
33 15.97 3.35 1. ... Kg6 2. Kg3 Kxg5 3. Kxh3 Kf4
4. Kg2 b5 5. Kf1 Kf3 6. Ke1 Ke4 7.
Ke2 Kf4 8. f3 Kf5 9. Ke3 Ke5 10. f4+
Kd5 11. Kf3 Kc5 12. Ke4 Kc6 13. Ke5
Kd7 14. f5 Ke7 15. f6+ Kd7 16. Kf5
Kd6 17. Kg6 Ke6 18. f7 Ke7
I stopped it after 16 seconds. I just selected random points to show the time and the eval climbing. This doesn't seem very hard, so no explanation why some would have trouble...
15-> 0.02 0.45 1. ... Kg6 2. Kg3 Kxg5 3. Kxh3 Kf4
4. Kg2 Kg4 5. f3+ Kf4 6. Kf2 b5 7.
Ke2 Ke5 8. Ke3 Kf5 9. Kd4
19 0.06 0.91 1. ... Kg6 2. Kg3 Kxg5 3. Kxh3 Kf4
4. Kg2 b5 5. Kf1 Kf3 6. Ke1 Ke4 7.
Ke2 Kf4 8. f3 Kf5 9. Ke3 Ke5 10. f4+
Kf6 11. Ke4
23 0.17 1.09 1. ... Kg6 2. Kg3 Kxg5 3. Kxh3 Kf4
4. Kg2 b5 5. Kf1 Kf3 6. Ke1 Ke4 7.
Ke2 Kf4 8. f3 Kf5 9. Ke3 Ke5 10. f4+
Kd5 11. Kf3 Ke6 12. Ke4 Kf6 13. Kd5
25 0.34 1.72 1. ... Kg6 2. Kg3 Kxg5 3. Kxh3 Kf4
4. Kg2 b5 5. Kf1 Kf3 6. Ke1 Ke4 7.
Ke2 Kf4 8. f3 Kf5 9. Ke3 Ke5 10. f4+
Kd5 11. Kf3 Kd6 12. Ke4 Ke6 13. f5+
Kf6 14. Kf4 Kf7
30 3.58 2.52 1. ... Kg6 2. Kg3 Kxg5 3. Kxh3 Kf4
4. Kg2 b5 5. Kf1 Kf3 6. Ke1 Ke4 7.
Ke2 Kf4 8. f3 Kf5 9. Ke3 Ke5 10. f4+
Kd5 11. Kf3 Kd6 12. Ke4 Ke6 13. f5+
Kf6 14. Kf4 Kf7 15. Ke5 Ke7 16. f6+
Kd7 17. Kf5 Kd6
33 15.97 3.35 1. ... Kg6 2. Kg3 Kxg5 3. Kxh3 Kf4
4. Kg2 b5 5. Kf1 Kf3 6. Ke1 Ke4 7.
Ke2 Kf4 8. f3 Kf5 9. Ke3 Ke5 10. f4+
Kd5 11. Kf3 Kc5 12. Ke4 Kc6 13. Ke5
Kd7 14. f5 Ke7 15. f6+ Kd7 16. Kf5
Kd6 17. Kg6 Ke6 18. f7 Ke7
I stopped it after 16 seconds. I just selected random points to show the time and the eval climbing. This doesn't seem very hard, so no explanation why some would have trouble...
-
Michel
- Posts: 2272
- Joined: Mon Sep 29, 2008 1:50 am
Re: Stupid engines
In the second position the evaluation probably takes into account that+1.4 for an extra doubled passer supported by King seems a gross underestimate (in KPPK this would be a guaranteed win), but for the second one, with the black King outside the passer's square, ('unstoppable passer') it is outright ridiculous.
black can go after the a pawn. It takes a bit of search to see that black
cannot stalemate himself.
-
Sven
- Posts: 4052
- Joined: Thu May 15, 2008 9:57 pm
- Location: Berlin, Germany
- Full name: Sven Schüle
Re: Stupid engines
It also takes a bit of search to find that 1.Kc4 Kb2 2.Kb4 Kc2 3.c4 is an easy win (although not the shortest one) while the 1.c4 Kb2 2.c5 Kxa3 3.c6 (much better 3.Kc3!) Kb2 ends up in an ending "queen vs. rook pawn a2" where many cases are drawn (this one isn't since the wK is close enough). That is "easy" when looking on the board as a human, but a static eval deep down in the tree will most probably not detect that.Michel wrote:In the second position the evaluation probably takes into account that+1.4 for an extra doubled passer supported by King seems a gross underestimate (in KPPK this would be a guaranteed win), but for the second one, with the black King outside the passer's square, ('unstoppable passer') it is outright ridiculous.
black can go after the a pawn. It takes a bit of search to see that black
cannot stalemate himself.
Evaluating this second position based on the "unstoppable passer" property might be misplaced, at least.
Sven
-
Eelco de Groot
- Posts: 4567
- Joined: Sun Mar 12, 2006 2:40 am
- Full name:
Re: Stupid engines
A late version of Fruit supposedly with added by Vas endgame knowledge that was removed in later versions, still has the same problem. Maybe an idea is to post this at the Rybka website and ask which is the first version that quickly solves this, including beta engines. Maybe then we would find the first version that is relatively free of any Fruit influences. But it is not my intention to turn this in yet another flamewarhgm wrote:Well, Fruit 2.1 is not generally known for being buggy, but there surely seems something wrong. (Although I was told The King also has trouble with this position.) Fruit takes 19 min to reach 33ply here, and even then it is only at -1.40 for black:
The weird thing is that when I follow the PVs that score more that 1, it ends in positions likeCode: Select all
33 -1.40 1247.2M 19:19.99 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Ke6 Ke4 Kf6 Kd4 Kf5 Ke3 Ke6 f5+ Kf6 Kd4 Kf7 Kc5 Kf6 Kxb5 Kxf5 Kxc4 33 -1.40 1232.7M 19:02.48 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Ke6 Ke4 Kf6 Kd4 Kf5 Ke3 Ke6 f5+ Kf6 Kd4 Kf7 Kc5 Kf6 Kxb5 Kxf5 Kxc4 32 -1.40 1021.1M 14:44.25 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 f5 Ke5 f6 Kxf6 Kd4 Kf5 Kc5 Ke4 Kxb5 Kd5 Kb4 Kd6 Kxc4 31 -1.26 486.7M 6:58.04 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Kd6 Ke4 Ke6 f5+ Kf6 Kd5 Kxf5 Kc5 Ke4 Kxb5 Ke3 Kxc4 Kd2 Kd4 Kxc2 30 -0.92 173.7M 2:11.04 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 f5 Ke5 f6 Kxf6 Kd4 Ke7 Kc5 Ke6 Kxb5 Kd5 Kxa4 29 -0.92 159.9M 1:54.96 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 f5 Ke5 f6 Kxf6 Kd4 Ke7 Kc5 Ke6 Kxb5 Kd5 Kxa4 28 -0.70 35.3M 0:28.09 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Kd6 Ke4 Ke6 f5+ Kd6 Kd4 Kc6 Ke3 Kd5 27 -0.64 28.6M 0:22.25 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Kd6 Ke4 Ke6 f5+ Kf6 Kf4 Kg7 Ke3 Kf6 26 -0.71 26.1M 0:19.97 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Kd6 Ke4 Ke6 f5+ Kf6 Kf4 Ke7 25 -0.28 7.0M 0:05.18 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kg4 Ke2 Kf4 f3 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Ke6 Kf2 Kd6 Kg3 Ke6 Kg4 24 -0.32 6.1M 0:04.38 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kd5 Kf2 Kd6 Kg3 Ke6 Kf4 Kd5 Kf5 Kd6 Ke4 Ke6 23 -0.31 3.9M 0:02.88 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kd5 Kf2 Kd6 f4 Ke6 Ke2 Kf6 Kd2 Kf5 Ke3 22 -0.21 1.3M 0:01.17 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kd5 Kf2 Kd6 f4 Ke6 Ke2 Kf5 Kf3 Ke6 21 -0.21 1.2M 0:01.02 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kd5 Kf3 Ke6 20 -0.19 596134 0:00.59 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Kf5 Ke3 Ke5 f4+ Kf5 Kf3 19 -0.19 484749 0:00.49 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Kf5 Ke3 Ke5 f4+ Kf5 Kf3 18 -0.08 243124 0:00.31 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Kf4 Ke2 Ke4 f3+ Kf4 Kf2 Ke5 Ke3 Kd5 17 -0.08 186221 0:00.26 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Ke5 Ke3 Kd5 16 -0.08 147100 0:00.22 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Kf3 Ke1 Ke4 Ke2 Kf4 f3 Ke5 Ke3 15 -0.18 107571 0:00.19 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 f3 Ke5 Kg3 Kf5 f4 Ke4 Kg4 Kd5 14 -0.16 72543 0:00.15 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 f3 Ke5 Kg3 Kf5 f4 Ke4 Kg4 13 0.00 46988 0:00.10 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 f3 Ke5 Kf2 Kf4 Kg2 12 -0.08 32740 0:00.09 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 f3 Ke5 Kf2 Kd5 Ke3 11 +0.16 21116 0:00.08 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 f3 Ke5 Kg3 Kd5 10 +0.04 13981 0:00.06 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 Kf1 Ke5 Ke2 9 +0.28 8457 0:00.02 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 b5 f3 Ke5 8 +0.16 5112 0:00.01 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 Ke4 Kg3 7 +0.28 2929 0:00.01 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 Ke4 6 +0.16 1595 0:00.00 Kg6 Kg3 Kxg5 Kxh3 Kf4 Kg2 5 +0.28 755 0:00.00 Kg6 Kg3 h2 Kxh2 Kxg5 4 +0.28 391 0:00.00 Kg6 Kg3 h2 Kxh2 Kxg5 3 +0.28 250 0:00.00 Kg6 Kg3 h2 Kxh2 Kxg5 3 +0.24 206 0:00.00 h2 Kg2 Kg6 f4 2 +0.28 87 0:00.00 h2 Kg2 Kg6 Kxh2 Kxg5 2 +0.16 35 0:00.00 Kg6 Kg3 Kxg5 Kxh3 1 +0.26 14 0:00.00 Kg6
[d]8/8/8/5k2/p1K5/P1P5/2P5/8 b
or
[d]8/8/8/8/p2K4/P1P5/2k5/8 w
Fruit does not do hash cuts in the PV, so the latter should always lead to the score-determining position. +1.4 for an extra doubled passer supported by King seems a gross underestimate (in KPPK this would be a guaranteed win), but for the second one, with the black King outside the passer's square, ('unstoppable passer') it is outright ridiculous.
Perhaps 16MB was is a bit on the low side, although in Pawn endings often very little hash is used. But that would explain the long times, but not the weird scoring.
I tried a Toga version that was I think around the same time as Rybka 2.2n2, that has similar behaviour, it is not some simple bug. Not implying the two are anywhere related. But it is really not my intention to turn this in yet another flamewar Please choose another reference engine or ask Fabien to program a new one 
Eelco
8/6k1/1p6/6P1/p1p5/P1P2K1p/2P2P2/8 b - -
Engine: Rybka 2.2n2 mp 32-bit PVtips5menbases (48 MB)
by Vasik Rajlich
2.00 0:00 -0.24 1...Kg6 2.Kg3 (81) 2
3.00 0:00 -0.13 1...Kg6 2.Kg3 (100) 3
4.00 0:00 -0.24 1...Kg6 2.Kg3 (132) 4
5.00 0:00 -0.13 1...Kg6 2.Kg3 (180) 5
6.00 0:00 -0.24 1...Kg6 2.Kg3 (252) 8
7.00 0:00 -0.23 1...Kg6 2.Kg3 (367) 7
8.00 0:00 -0.33 1...Kg6 2.Kg3 (566) 12
9.00 0:00 -0.32 1...Kg6 2.Kg3 (821) 17
10.01 0:00 -0.32 1...Kg6 2.Kg3 Kxg5 3.Kxh3 (1.395) 30
11.01 0:00 -0.43 1...Kg6 2.Kg3 Kxg5 3.Kxh3 (2.030) 32
12.01 0:00 -0.45 1...Kg6 2.Kg3 Kxg5 3.Kxh3 (2.997) 48
13.01 0:00 -0.43 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (4.232) 54
14.01 0:00 -0.49 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (6.356) 69
15.01 0:00 -0.59 1...Kg6 2.Kg3 Kxg5 3.f3 h2 4.Kxh2 Kf4 (11.177) 81
16.01 0:00 -0.68 1...Kg6 2.Kg3 Kxg5 3.f3 h2 4.Kxh2 Kf4 (16.727) 91
17.01 0:00 -0.68 1...Kg6 2.Kg3 Kxg5 3.f3 h2 4.Kxh2 Kf4 (19.998) 100
18.01 0:00 -0.83 1...Kg6 2.Kg3 Kxg5 3.f3 h2 4.Kxh2 Kf4 (34.224) 117
19.01 0:00 -0.76 1...Kg6 2.Kg3 Kxg5 3.f3 h2 4.Kxh2 Kf4 (39.885) 118
20.01 0:00 -0.94 1...Kg6 2.Kg3 Kxg5 3.f3 h2 4.Kxh2 Kf4 (66.008) 130
21.01 0:00 -1.05 1...Kg6 2.Kg3 Kxg5 3.f3 h2 4.Kxh2 Kf4 (108.318) 136
22.01 0:01 -1.06 1...Kg6 2.Kg3 Kxg5 3.f3 h2 4.Kxh2 Kf4 (160.307) 136
23.01 0:02 -1.05 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (274.987) 139
24.01 0:02 -1.05 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (385.864) 138
25.01 0:03 -1.04 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (483.893) 140
26.01 0:04 -1.04 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (546.956) 137
27.01 0:20 -1.06 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (2.638.014) 133
28.01 0:57 -1.13 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (7.219.936) 129
29.01 0:58 -1.13 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (7.417.097) 129
30.01 3:18 -1.33 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (24.924.348) 128
31.01 3:21 -1.33 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (25.258.656) 128
32.01 12:23 -1.83 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (87.345.780) 120
33.01 21:55 -2.54 1...Kg6 2.Kg3 Kxg5 3.Kxh3 Kf4 (183.269.926) 142
best move: Kg7-g6 time: 30:48.797 min n/s: 265.317.814 nodes: 265.317.814
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan
-
Desperado
- Posts: 879
- Joined: Mon Dec 15, 2008 11:45 am
Re: Stupid engines
Well,...
8/8/8/5k2/p1K5/P1P5/2P5/8 b
8/8/8/8/p2K4/P1P5/2k5/8 w
I dont see any problems here, even unstoppable passer property isnt a problem. nor it has sth do with passed pawn extension or similar stuff.
I think this problem is caused by hashtable replacement scheme used by
Fruit. A while back i were able to see such behaviour in my engine.
The reason was the "depth" replacement condition for already existing
entries. Some entries get a super high draft which locks the slot so to say,
while it is containing 1.xx value for example. This may be one reason !?.
The solution was trivial, always replacing an existing entry...
@HG : maybe you like to report what happens in Fruit, if you always replace an entry which already exists...
Michael
8/8/8/5k2/p1K5/P1P5/2P5/8 b
Code: Select all
Analysis by Nemo SP64 0.51 00:
1...Ke6 2.Kb5 Kd6 3.c4 Kc7 4.c5 Kb7 5.c6+ Kc8 6.Kb6 Kb8 7.c7+ Kc8 8.c4 Kd7 9.Kb7 Kd6 10.c8D Ke5 11.c5 Kd4 12.c6 Kc3 13.c7 Kb2 14.Dh8+ Kc1 15.c8D+ Kd2 16.Dcc3+ Ke2 17.Dh2+
+- (#17) Tiefe: 26 00:00:09 32823kN
(H, privat 06.10.2011)
Code: Select all
Analysis by Nemo SP64 0.51 00:
1.c4 Kd2 2.c5 Ke1 3.c6 Kf2 4.c7 Kg3 5.c8D Kf4 6.De6 Kf3 7.De4+ Kg3 8.Ke3 Kh2 9.Kf3 Kh1 10.Dc2 Kg1 11.Dg2#
+- (#11) Tiefe: 18 00:00:00 1484kN
(H, privat 06.10.2011)
I think this problem is caused by hashtable replacement scheme used by
Fruit. A while back i were able to see such behaviour in my engine.
The reason was the "depth" replacement condition for already existing
entries. Some entries get a super high draft which locks the slot so to say,
while it is containing 1.xx value for example. This may be one reason !?.
The solution was trivial, always replacing an existing entry...
@HG : maybe you like to report what happens in Fruit, if you always replace an entry which already exists...
Michael