Hard-Talkchess-2019 set

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

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Vinvin
Posts: 4390
Joined: Thu Mar 09, 2006 8:40 am
Full name: Vincent Lejeune

Re: Hard-Talkchess-2019 set

Post by Vinvin » Wed Sep 25, 2019 5:32 pm

lech wrote:
Wed Sep 25, 2019 4:48 pm
Vinvin wrote:
Tue Sep 24, 2019 5:25 pm
Vinvin wrote:
Mon Sep 23, 2019 9:21 pm
...
And I'm adding 6 older positions, sometimes easy and sometimes hard :

Code: Select all

...
rk1b4/p2p2p1/1P6/2R2P2/8/2K5/8/5B2 w - - bm Rc8+;
...
Now I remember why this position is not good : Rc8+ is sometime evaluated as 0.00 as some other moves.
I'm removing this one.
Vincent, are you sure?
This position seems to be a very good winning one. :D
Good luck!
Yes, Rc8+, Ba6 or Rd5 can be chosen randomly but only one is winning.

Code: Select all

Stockfish_19091615_x64_modern:
...
 56/44	04:50	2 191 168 105	7 546 072	 0,00	1.Ba6 axb6 2.Rc8+ Ka7 3.Rxa8+ Kxa8 4.Bc8 Kb8 5.Bxd7 Bf6+ 6.Kb3 Kc7 7.Bb5 Kd6 8.Kb4 Ke5 9.Bd3 Kd5 10.Bc2 Kd6 11.Kb5 Bd4 12.Kc4 Bf2 13.Bd3 Bc5 14.Kb5 Kd5 15.Bb1 Ke5 16.Bc2 Kd5
 56/66	04:50	2 191 168 105	7 546 072	 0,00	1.Rc8+ Kxc8 2.Ba6+ Kb8 3.b7 Bf6+ 4.Kc4 Be5 5.Kd5 Bg3 6.Kc4
 56/23	04:50	2 191 168 105	7 546 072	 0,00	1.Rd5 Kb7 2.Rd6 Rc8+ 3.Kb3 Bxb6 4.Rxd7+ Rc7 5.Rd6 Rf7 6.Bc4 Rxf5 7.Rd7+ Kc6 8.Rxg7 Rf3+ 9.Ka4 Kc5 10.Rg4 Re3 11.Bb3 Rxb3
---------------------------------------------------------------------------
 57/36	06:15	2 872 072 839	7 641 214	 0,00	1.Ba6 axb6 2.Rc8+ Ka7 3.Rxa8+ Kxa8 4.Bc8 Kb8 5.Bxd7 Bf6+ 6.Kc4 Kc7 7.Bb5 Kd6 8.Ba4 Be5 9.Bc2 Bg3 10.Bd3 Ke5 11.Bc2 Bf2 12.Kb5 Bc5 13.Bb1 Bd4 14.Kc6 Be3 15.Bc2 Bd4
 57/15	06:15	2 872 072 839	7 641 214	 0,00	1.Rc8+ Kxc8 2.Ba6+ Kb8 3.b7 Bf6+ 4.Kc4 Be5 5.Kd5 Bg3 6.Kc4 Bh2 7.Kd5 Bg3
 57/31	06:15	2 872 072 839	7 641 214	 0,00	1.Rd5 Kb7 2.Rd6 Rc8+ 3.Kd3 Bxb6 4.Rg6 Rf8 5.Rxg7 Kc6 6.Rg6+ Kc5 7.f6 Bd8 8.Bh3 Bxf6 9.Bxd7 Kd6 10.Bg4 a5 11.Kc4 a4 12.Bd1 a3 13.Kb3 Kc5 14.Kxa3

Dann Corbit
Posts: 10178
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Hard-Talkchess-2019 set

Post by Dann Corbit » Wed Sep 25, 2019 7:12 pm

I think what he refers to is that engines find the right move, but score it a draw and not a checkmate.
Automated systems that analyze the output say that the engine chose correctly, when really it might draw the game because it only found the draw path.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

Paloma
Posts: 855
Joined: Thu Dec 25, 2008 8:07 pm

Re: Hard-Talkchess-2019 set

Post by Paloma » Wed Sep 25, 2019 7:17 pm

Code: Select all

rk1b4/p2p2p1/1P6/2R2P2/8/2K5/8/5B2 w - - 0 1

Analysis by Sting 17 (based on Stockfish 2.1.1) 64bit:

...
1.Rc8+ Kb7 2.Bg2+ Kxb6 3.Rxa8 Bg5 4.Kc4 a5 5.Kd5 Kb5 6.Rg8 Bf6 7.Rb8+ Ka4 8.Kd6 Ka3 9.Bd5 Ka4 10.Kxd7 Bc3 11.Ke8 Be5 12.Rb7 Bf6 13.Kf7 Bc3 14.Kg6 Be5 15.Bg2 Ka3 16.Rxg7 Kb4 17.f6 Bd6 18.f7 Kb3 19.Rg8 Bc5 20.f8Q Bd4 21.Rg7 Bg1 22.Rg8 Kc2 
  Weiß steht klar auf Gewinn: +- (76.56)  Tiefe: 42   00:14:12  5833MN
1.Rc8+ Kb7 2.Bg2+ Kxb6 3.Rxa8 Bg5 4.Kc4 Kc7 5.Kd5 Kb7 6.Kd6+ Ka6 7.Bf1+ Kb7 8.Rg8 Bh6 9.Rd8 Bf4+ 10.Kd5 Kc7 11.Ra8 Bc1 12.Bb5 Kb7 13.Rg8 Bh6 14.Re8 Bg5 15.Rf8 a6 16.Bf1 a5 17.Bb5 a4 18.Bxa4 Ka6 19.Rf7 Ka5 20.Rxd7 Kb4 21.Rxg7 Bd8 22.Bb5 Bf6 
  Weiß steht klar auf Gewinn: +- (79.87)  Tiefe: 43   00:14:44  6037MN
1.Rc8+ Kb7 2.Bg2+ Kxb6 3.Rxa8 Bg5 4.Kc4 Kc7 5.Kd5 Kb7 6.Kd6+ Ka6 7.Bf1+ Kb7 8.Rg8 Bh6 9.Rd8 Bf4+ 10.Kd5 Kc7 11.Ra8 Bg5 12.Bb5 Kb7 13.Re8 Kc7 14.Rg8 Be7 15.Bf1 Bf6 16.Bg2 a5 17.Bf1 a4 18.Ra8 a3 19.Rxa3 Kd8 20.Re3 Bh4 21.Re4 Bf2 22.Bb5 Ba7 23.Rg4 Ke7 24.Rg6 
  Weiß steht klar auf Gewinn: +- (79.87)  Tiefe: 44   00:15:49  6549MN

(25.09.2019)

Dann Corbit
Posts: 10178
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Hard-Talkchess-2019 set

Post by Dann Corbit » Wed Sep 25, 2019 8:01 pm

Yes, some engines will find the mate score.
Others will find the key move and give a score of 0 and a drawing plan.
The GUI will say, +1, you got the right answer.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

MikeB
Posts: 3532
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Hard-Talkchess-2019 set

Post by MikeB » Thu Sep 26, 2019 1:16 am

Vinvin wrote:
Tue Sep 24, 2019 5:25 pm
Vinvin wrote:
Mon Sep 23, 2019 9:21 pm
...
And I'm adding 6 older positions, sometimes easy and sometimes hard :

Code: Select all

...
rk1b4/p2p2p1/1P6/2R2P2/8/2K5/8/5B2 w - - bm Rc8+;
...
Now I remember why this position is not good : Rc8+ is sometime evaluated as 0.00 as some other moves.
I'm removing this one.
i would leave it in - excellent position = only Rc8! wins - but it has to be shown in the score too.

dep score nodes time (not shown: tbhits knps seldep)
43 +100.00 704.2M 0:24.93 Rc8+ Kb7 Bg2+ Kxb6 Rxa8 Bg5 Kc4 a5 Bf1 Bd2 Rg8 Bh6 Kd5 Kc7 Bb5 d6 Ra8 Kb6 Bf1 Bf4 Ra6+ Kc7 Rxa5 Kb6 Ra6+ Kc7 Ra7+ Kd8 Rxg7 Be5 f6 Bf4 f7 Ke7 f8=Q+ Kxf8

Paloma
Posts: 855
Joined: Thu Dec 25, 2008 8:07 pm

Re: Hard-Talkchess-2019 set

Post by Paloma » Thu Sep 26, 2019 5:05 pm

Yes, it's a mate in 28!

1.Rc8+ Kb7! 2.Bg2+ Kxb6 3.Rxa8 Bg5 4.Kd4 Bh4 5.Kd5 (24#) Be7 6.Rg8 Bf6 7.Rf8! (22#) Ba1 8.Bf1 and so on.
Solution backwards with Houdini 6.02, 3core, 8192MB Hash

1...Kxc8? 2.Ba6+ (22#) Kb8 3.b7 Bb6 4.Kc4 (20#) Bc7 5.Kd5 Bg3 6.f6 gxf6 7.Ke4
______________________or 3...Bxf6+ 4.Kc4 (20#) Be7 5.Kd5 Ba3 6.Ke5

Kubbel=L (+0440.23c3b8), Shakhmaty v SSSR#760, 1936
.
.

Vinvin
Posts: 4390
Joined: Thu Mar 09, 2006 8:40 am
Full name: Vincent Lejeune

Re: Hard-Talkchess-2019 set

Post by Vinvin » Thu Sep 26, 2019 7:00 pm

MikeB wrote:
Thu Sep 26, 2019 1:16 am
Vinvin wrote:
Tue Sep 24, 2019 5:25 pm
Vinvin wrote:
Mon Sep 23, 2019 9:21 pm
...
And I'm adding 6 older positions, sometimes easy and sometimes hard :

Code: Select all

...
rk1b4/p2p2p1/1P6/2R2P2/8/2K5/8/5B2 w - - bm Rc8+;
...
Now I remember why this position is not good : Rc8+ is sometime evaluated as 0.00 as some other moves.
I'm removing this one.
i would leave it in - excellent position = only Rc8! wins - but it has to be shown in the score too.
...
But I don't want to have to do manual checking.

Dann Corbit
Posts: 10178
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Hard-Talkchess-2019 set

Post by Dann Corbit » Thu Sep 26, 2019 10:29 pm

Vinvin wrote:
Thu Sep 26, 2019 7:00 pm
But I don't want to have to do manual checking.
And neither does anyone else running the test suites.

It is similar with a couple positions from the sting suite, which require specific outcomes.

Chess GUIs do not check for these things.

Sure, we can check by hand.

What if we are running 30 engines against a suite? It gets more and more problematic.

Now, the position is still nice and interesting. But it tends to generate false positives.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

Vinvin
Posts: 4390
Joined: Thu Mar 09, 2006 8:40 am
Full name: Vincent Lejeune

Re: Hard-Talkchess-2019 set

Post by Vinvin » Tue Oct 01, 2019 6:36 pm

Vinvin wrote:
Tue Sep 24, 2019 12:32 pm
...
This will be Beta3.

I plan to run Stockfish (15 minutes, 6 cores) 3 or 4 times. Most of this is already done.
+ I'll run Houdini 1.5 and Critter 1.6a later.

One more idea is to create studies on Lichess for all the good and the bad test positions.
For what I know, there's no tool to find a position on Lichess, so we have to make link between both here, on the forum.
Follow-up here : viewtopic.php?p=813987

Post Reply