the "Black Hole" generator ("AI")

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

lech
Posts: 1175
Joined: Sun Feb 14, 2010 10:02 pm

the "Black Hole" generator ("AI")

Post by lech »

A simple example for the "black hole" generator.

[d] 8/1P6/8/1b1K4/6p1/5pP1/5P1B/7k w - -

White can promote to Queen with a huge advantage. But 1. b8Q Bf1 2. Qb1 Kxh2 3. Qxf1 leads to stalemate.
If White does not capture the black Bishop, this one goes to h3 and it is a fortrees.
For this, White has to promote to Rook to avoid stalemate.

Sting SF 25 [Hash = 1024M, Threads = 1]
.............................
info depth 23 multipv 1 score cp 989 nodes 1346370 nps 363588 time 3703 pv b7b8q b5f1 b8b4 f1h3
.............................
info depth 31 multipv 1 score cp 989 nodes 17435536 nps 529858 time 32906 pv b7b8q b5f1 b8b4 f1h3

No gain! The generator turns on and cp (centi pawn) goes to ZERO.
info depth 32 multipv 1 score cp 981 upperbound nodes 20520776 nps 312402 time 65687 pv b7b8q b5f1
.............................
info depth 32 multipv 1 score cp 380 upperbound nodes 78266836 nps 483221 time 161969 pv b7b8q b5f1

The correct first move appears.
info depth 32 multipv 1 score cp 193 nodes 247461861 nps 509557 time 485641 pv b7b8r b5d3 b8g8 d3f5 d5e5 f5d7 e5f4 h1g2 h2g1 g2h3 g8h8 h3g2 h8h2 g2f1 h2h7 d7e6h7g7 e6b3 g7g4 f1e2 g4g7 e2d3 g7d7 d3c3 d7d2 c3d2 f4f3 d2d3 g3g4 d3d4 f3f4 b3c4f4f5 d4c5 f2f4 c5d5

But still it goes down.
info depth 33 multipv 1 score cp 181 upperbound nodes 257692480 nps 485496 time 530781 pv b7b8r b5d3
.............................
info depth 33 multipv 1 score cp 22 upperbound nodes 410726811 nps 515624 time 796562 pv b7b8r b5a6
info depth 33 multipv 1 score cp 0 nodes 464157662 nps 518385 time 895391 pv b7b8r b5a6 d5e4 h1h2 b8b6 a6c8 b6b3 h2g2 e4e3 g2h3 b3b6 c8d7 e3d4 h3g2 b6b7 d7c8 b7c7 c8f5 d4e3 f5e6 c7e7 e6c8 e7e8 c8d7 e8e7 d7c8

Cp = 0 and now the generator try to help to escape from zero (black hole).
info depth 34 multipv 1 score cp 8 lowerbound nodes 481317184 nps 514941 time 934703 pv b7b8r
info depth 34 multipv 1 score cp 0 nodes 513114743 nps 517684 time 991172 pv b7b8r b5a6 d5e4 h1h2 e4f4 h2g2 f4g4 g2f2 g4f4 f2g2 b8b2 f3f2 f4e3 g2g3 b2f2 a6b5 f2f6 g3g4 f6d6 b5e8 e3f2 g4g5 f2g1 e8f7 g1h2 g5f5 d6b6 f7h5 h2h3 f5e4 h3g3 e4f5 b6b5 f5g6 g3f2
.............................
info depth 43 multipv 1 score cp 8 lowerbound nodes 1405674332 nps 534242 time 2631156 pv b7b8r
info depth 43 multipv 1 score cp 0 nodes 1607468296 nps 538312 time 2986125 pv b
7b8r b5a6 d5e4 h1h2 e4f4 h2g2 f4g4 g2f2 g4f4 f2g2 b8b2 f3f2 f4e3 g2g3 b2f2 g3g4 f2f8 a6b7 e3d4 b7g2 d4c5 g2e4 c5b6 e4f5 b6c5 g4f4 f8h8 f5g4 c5b6 f4e5 b6a7 e5f4 a7b8 f4e5 h8h4 g4f5 h4b4 e5e6 b4h4 e6f7 h4f4 f7f6 f4d4

Finally, it is done.
info depth 44 multipv 1 score cp 8 lowerbound nodes 1750042345 nps 528970 time 3308391 pv b7b8r
info depth 44 multipv 1 score cp 16 lowerbound nodes 4212151550 nps 842855 time 4997477 pv b7b8r
.............................
info depth 44 multipv 1 score cp 175 lowerbound nodes 4293647886 nps 841176 time 5104337 pv b7b8r
info depth 44 multipv 1 score cp 267 lowerbound nodes 4295222121 nps 841135 time 5106458 pv b7b8r

info depth 44 multipv 1 score cp 319 nodes 4376931676 nps 838859 time 5217717 pv b7b8r b5a6 b8b2 h1h2 d5d4 a6e2 b2b1 e2a6 b1b8 h2g2 b8b2 g2h2 d4e5 h2g2 e5e4 a6c8 e4f4 c8d7 f4e3 d7e6 b2d2 e6c8 d2c2 c8e6 c2c5 e6f7 c5e5 g2h2 e5e4 h2h3 e4e7 f7b3 e7h7 h3g2 h7c7 b3d5

It was a simple example. Much more difficult is the next position with two Rook promotions:
[pgn][FEN "4q1bk/PP4p1/1p4P1/8/1p3QPP/2Pp4/3P1pP1/K7 w - - 0 1"] 1. a8R Qe1 2. Kb2 bxc3 3. Kxc3 Qc1 4. Kd4 Qc4 5. Ke3 Qxf4 6. Kf4 f1Q 7. Kg5 Qf6 8. Kh5 Qe5 9. g5 Qe2 10. g4 Qf3 11. b8R [/pgn]

Sting 25 gets it too!
Maybe, I can't be friendly, but let me be useful.
Paloma
Posts: 1219
Joined: Thu Dec 25, 2008 9:07 pm
Full name: Herbert L

Re: the "Black Hole" generator ("AI")

Post by Paloma »

>time 5217717 pv b7b8r

nearly 1hour and 45 min. :shock:
Pi4Chess
Posts: 253
Joined: Mon Nov 16, 2020 12:13 pm
Full name: Manuel Rivera

Re: the "Black Hole" generator ("AI")

Post by Pi4Chess »

Very interesting :shock:
Pi4Chess
Posts: 253
Joined: Mon Nov 16, 2020 12:13 pm
Full name: Manuel Rivera

Re: the "Black Hole" generator ("AI")

Post by Pi4Chess »

How do you know it enters in "the black hole generator" ? (is the "upperbound" string and "lowerbound" string the indication ?)

You launched a classic analisys search ? or which command did you use for this position ?

Thanks for your insights
Pi4Chess
Posts: 253
Joined: Mon Nov 16, 2020 12:13 pm
Full name: Manuel Rivera

Re: the "Black Hole" generator ("AI")

Post by Pi4Chess »

Paloma wrote: Mon Dec 21, 2020 11:37 pm >time 5217717 pv b7b8r

nearly 1hour and 45 min. :shock:
1h and 26 minutes ;-)
Pi4Chess
Posts: 253
Joined: Mon Nov 16, 2020 12:13 pm
Full name: Manuel Rivera

Re: the "Black Hole" generator ("AI")

Post by Pi4Chess »

For the first position, Crystal finds b7-b8R in less than 1 second and Cfish in 1m25sec.

For the second position it's black to play and win ?
Pi4Chess
Posts: 253
Joined: Mon Nov 16, 2020 12:13 pm
Full name: Manuel Rivera

Re: the "Black Hole" generator ("AI")

Post by Pi4Chess »

Oh ok i figured out.
Jouni
Posts: 3791
Joined: Wed Mar 08, 2006 8:15 pm
Full name: Jouni Uski

Re: the "Black Hole" generator ("AI")

Post by Jouni »

Analysis by Crystal 110920:

...
1.a8R Qe1+ 2.Kb2 bxc3+ 3.Kxc3 Qc1+ 4.Kd4 Qc4+ 5.Ke3 Qxf4+ 6.Kxf4 f1Q+ 7.Kg5 Qf6+ 8.Kh5 Qe5+ 9.g5 Qe2+ 10.g4 Qf3 11.b8R Qd5 12.Rd8 Qc4 13.Ra3 b5 14.Rc3 Qe6 15.Rcxd3 b4 16.Re3 Qd5 17.Rxd5 Bxd5 18.Re8+ Bg8 19.Rb8 b3 20.d4 b2 21.Rxb2 Bd5 22.Rb8+ Bg8 23.Rb7 Be6 24.Re7 Bb3 25.Rd7 Be6 26.Rd8+ Bg8 27.Rd6
+- (54.67) Depth: 27/72 00:00:28 384mN, tb=193929
...
Jouni
lech
Posts: 1175
Joined: Sun Feb 14, 2010 10:02 pm

Re: the "Black Hole" generator ("AI")

Post by lech »

At first, time usded by Sting it is a reason of my very old and slow computer mainly.
I never use a special commands to run searching of Sting (no cheating!).
I like that other engines (CFish, Crystal ...) try to make engines "normal" too.
I hope that not only Sting is able to solve 100/100 positions of my "IQ" test without cheating - "one brain in one head".
With some forward-backward (or special parameters) tiricks each engine is able to solve each position.
The newest version of "IQ" test is here: http://talkchess.com/forum3/viewtopic.p ... 91#p876557
Maybe, I can't be friendly, but let me be useful.
lech
Posts: 1175
Joined: Sun Feb 14, 2010 10:02 pm

Re: the "Black Hole" generator ("AI")

Post by lech »

But it is not the end of the "Black Hole" idea.
The first problem: One of many decisions is wrong but has huge advantage (described above).
The second problem: One of many decision is good but a success is far or out of range.
It needs an "intuition" and Sting 25 has it.

To prove it, I selected: nate in 74 - Popov & Khanyan.

[d] qn6/b1Kp3p/p1pB3p/p6p/P1NkP3/2pP4/2B5/8 w - -

Here searching has no sense, the solution is too long.
I started auto-playing (Sting vs. Sting; common hash) with only 2 seconds per move.
Sting 25 avoids draws ("Black Hole") and selects the all correct moves of the long plan only thanks to the "intuition".

Sting 25 (based on Stockfish 2.1.1) by Marek Kwiatkowski
setoption name hash value 1024
position fen qn6/b1Kp3p/p1pB3p/p6p/P1NkP3/2pP4/2B5/8 w - -
go movetime 2000 play
d6f4 d4c5 f4e3 c5b4 e3c1 b4c5 c1a3 c5d4 a3d6 h5h4 d6f4 d4c5 f4e3 c5b4 e3c1 b4c5 c1a3 c5d4 a3d6 h6h5 d6f4 d4c5 f4e3 c5b4 e3c1 b4c5 c1a3 c5d4 a3d6 h4h3 d6f4 d4c5 f4e3 c5b4 e3c1 b4c5 c1a3 c5d4 a3d6 h5h4 d6f4 d4c5 f4e3 c5b4 e3c1 b4c5 c1a3 c5d4 a3d6 h7h5 d6f4 d4c5 f4e3 c5b4 e3c1 b4c5 c1a3 c5d4 a3d6 h3h2 d6h2 d4c5 h2d6 c5d4 d6f4 d4c5 f4e3 c5b4 e3c1 b4c5 c1a3 c5d4 a3d6 h4h3 d6f4 d4c5 f4e3 c5b4 e3c1 b4c5 c1a3 c5d4 a3d6 h5h4 d6f4 d4c5 f4e3 c5b4 e3c1 b4c5 c1a3 c5d4 a3d6 h3h2 d6h2 d4c5
h2d6 c5d4 d6f4 d4c5 f4e3 c5b4 e3c1 b4c5 c1a3 c5d4 a3d6 h4h3 d6f4 d4c5 f4e3 c5b4 e3c1 b4c5 c1a3 c5d4 a3d6 h3h2 d6h2 d4c5 h2d6 c5d4 d6f4 d4c5 f4e3 c5b4 e3c1 b4c5 c1a3 c5d4 a3d6 a7b6 c7b6 a8a7 b6a7 c6c5 d6e5 Black is mated
Maybe, I can't be friendly, but let me be useful.