7 steps to heaven. 7 testpositions of variable difficulty!

[peter]

But even much more exiting I find the success in this study with a tool, that had been told about here:

http://www.talkchess.com/forum/viewtopi ... 62&t=47468

In this posting see also the links to the hp of the author and to the corresponding thread in Rybkaforum.

I told Albitex about your study, George, and with the latest version he just sent to me, Montecarlo Cockroach manages to find the solution within about half an hour, the longer it goes on searching, the better the result is getting, this is after 1h20':


___________

Montecarlo Cockroach v.1.0

Analysis of the Fri Mar 22 10:17:01 2013


You have planned this parameters:

FEN : 1k6/3p4/1B6/4Pp1p/1p5R/1p4p1/pP3n2/K6n w - - 0 1

CORE THREADS: 12

Log: yes Analisys: full

Depth: 13 Multipv:3

Delta=150

n.Moves LOW= 7

Depth LOW=13 Multipv LOW= 3

Delta LOW= 150

moves white:

1. Rh4:h1 n. games 278 n.win=5 n.draw=93 n.lose=180 p.max=18%
1. Rh4:b4 n. games 192 n.win=8 n.draw=14 n.lose=170 p.max=7%
1. Rh4 c4 n. games 171 n.win=3 n.draw=17 n.lose=151 p.max=6%
1. Bb6 e3 n. games 219 n.win=1 n.draw=11 n.lose=207 p.max=2%
1. Rh4:h5 n. games 233 n.win=1 n.draw=1 n.lose=231 p.max=0%

n.total games=1093 time: 1h 22' 8,667''

BTW, didn't you want to tell us about the origin of the study, George?

[ernest]

I find the success in this study
.......
1. Rh4:h1 n. games 278 n.win=5 n.draw=93 n.lose=180 p.max=18%
1. Rh4:b4 n. games 192 n.win=8 n.draw=14 n.lose=170 p.max=7%
1. Rh4 c4 n. games 171 n.win=3 n.draw=17 n.lose=151 p.max=6%
1. Bb6 e3 n. games 219 n.win=1 n.draw=11 n.lose=207 p.max=2%
1. Rh4:h5 n. games 233 n.win=1 n.draw=1 n.lose=231 p.max=0%

Hi Peter,

Why do you call this success?
Doesn't that only show that Rh4-h1 is a move worthy of further attention?

Or maybe you mean that to discover Rh4-h1 is interesting because it is such an improbable move?...

[peter]

Hi Ernest!

The principles of MonteCarlo- analysis cannot be compared to other methods directly.
I see it as success, if a tool brings up the in other ways not to be found solution as best move, which the upper example does.
Maybe you have seen my other "success" with Fritz-GUI's "deep position analysis", here the difference is not so big, there we can get only the right ranking of the candidate moves too and must not hope to see the numeric right eval given by the engines' output.
And then: do you know another way to get 1.Rxh1 as best move by any kind of engine- or GUI- support?
I tried IdEA too, but I didn't succeed neither.

In the meantime I got a much better result with MC Cockroach then the shown one, but a crash made it impossible to document it, there I had about 40 games remis with 1.Rxh1 with not a single loss counted, but still other candidate moves had some remis too, so it's still statistics only, and of course you can never be sure about a statistical result compared to a forced variant.

But what I like very specially with MC Cockroach, is the possibility, to adopt the parameters according to the position so well as for depth of computing and as for so called "limited" in contrast to "full" analysis, that to me it seems to be the missing link between a GUI's automated deep position analysis (which is in most cases an in between of MC and MV mode too) and Monte Carlo itself.
"Limited" means, the games don't have to be played out to the very end, you give a number of moves after which the engine's eval is compared to the others and that seems to me the most hopeful way of engine- assisted analyis of positions, engine cannot evaluate statically only or not exactly enough because of their horizon- effect.
Especially unclear postions of opening and early middlegame maybe could be analysed better with such methods than with engines- only- analysis or Monte Carlo- analyis only, you see?
As a matter of fact in opening theory even human masters have to rely on statistics of certain candidate moves to some extent.

The result with 40 remis to 0 losses after 1.Rxh1 I got with a depth of 21 moves instead of the 13 of the result of the posting.

Albitex is working very hard on his tool, as he tells me, and I'm rather curious about the advancements he's going to make further on

[Paloma]

•Very hard. If computers had the slightest ability of planning, then they would solve it in seconds.
1.Ke8! or 1.Ke7! wins! All other moves like e.g 1.Kxf7? or Kg7? etc only draw.
[d]5K2/k4p1p/5p1p/1p3p2/bP6/1p6/1P5p/7B w - - 0 1

[d]1K6/1B3p1p/1k3p1p/1p3p2/bP6/1p6/1P5p/8 b - - 0 6
Zugzwang! And then, after all black kingside pawns been captured, white wins the b5 pawn and bishop. But to deep even for the best Engines, more than 80 plies.

[lech]

•Very hard. If computers had the slightest ability of planning, then they would solve it in seconds.
1.Ke8! or 1.Ke7! wins! All other moves like e.g 1.Kxf7? or Kg7? etc only draw.
[d]5K2/k4p1p/5p1p/1p3p2/bP6/1p6/1P5p/7B w - - 0 1

[d]1K6/1B3p1p/1k3p1p/1p3p2/bP6/1p6/1P5p/8 b - - 0 6
Zugzwang! And then, after all black kingside pawns been captured, white wins the b5 pawn and bishop. But to deep even for the best Engines, more than 80 plies.

You are right, probably only without pawns f7 and h7 some engines are able to solve this now.

[George Tsavdaris]

Hi again, George!

May I at first ask you once again for the origin of the study?
You see, I just asked you twice already.

Sorry i just now saw it(and forgot about it the first time) but i just don't have the source of the position written anywhere.
It's an interesting study though, a complete anti-computer one.

[lech]

Sting-sf-3 (beta) gets this position, without ugly pawns h7 and f7 (only time consuming) on my dualcore in two minutes.
[d]5K2/k7/5p1p/1p3p2/bP6/1p6/1P5p/7B w - - 0 1 bm Ke8(e7);

Code: Select all

info depth 38 seldepth 36 multipv 1 score cp 0 nodes 62053328 nps 1496378 time 4
1469 pv f8e7 a7b8 e7f6 b8c7 f6f5 c7b6 f5g6 b6a7 g6h6 a7b6 h6h5 b6c7 h5g4 c7d6 g4
f4 d6e6 h1c6 e6d6 c6h1
info depth 39
info currmove f8e7 currmovenumber 1
info nodes 96058159 nps 1533079 time 62657
info currmove h1f3 currmovenumber 2
info currmove h1g2 currmovenumber 3
info currmove f8f7 currmovenumber 4
info currmove h1e4 currmovenumber 5
info currmove h1d5 currmovenumber 6
info currmove f8g7 currmovenumber 7
info nodes 100597578 nps 1534014 time 65578
info currmove h1c6 currmovenumber 8
info currmove h1a8 currmovenumber 9
info currmove h1b7 currmovenumber 10
info currmove f8g8 currmovenumber 11
info currmove f8e8 currmovenumber 12
info depth 39 seldepth 42 multipv 1 score cp 0 nodes 100599846 nps 1533674 time
65594 pv f8e7 a7b8 e7f6 b8c7 f6f5 c7b6 f5g6 b6a7 g6h6 a7b6 h6h5 b6c7 h5g4 c7d6 g
4f4 d6e6 h1c6 e6d6 c6h1
info depth 40
info currmove f8e7 currmovenumber 1
info depth 40 seldepth 0 multipv 1 score cp 8 lowerbound nodes 211917977 nps 158
0557 time 134078 pv f8e7 a7b8 e7d8 b8a7 d8d7 f5f4 d7c7 a7a6 h1b7 a6a7 b7g2 a7a6
g2f3 f6f5 f3b7 a6a7 b7g2 a7a6 g2h1 a6a7 h1b7 f4f3 b7f3 a7a6 f3h1 f5f4 h1b7 a6a7
b7f3 a7a6 f3h1 f4f3 h1f3 a6a7 f3b7 h2h1n b7h1
info nodes 211917977 nps 1580368 time 134094
info currmove f8e7 currmovenumber 1
info depth 40 seldepth 0 multipv 1 score cp 16 lowerbound nodes 213051055 nps 15
77419 time 135063 pv f8e7 a7b8 e7d8 f5f4 d8d7 f4f3 h1f3 h6h5 f3h1 h5h4 h1g2 h4h3
 g2f3 b8a7 d7c7 a7a6 f3b7 a6a7 b7h1 a7a6 h1f3 a6a7 f3b7 h2h1n b7h1 a7a6 c7d6 a6a
7 d6d7 a7b8 h1f3 b8a7 d7c7 a7a6 f3h1 a6a7 c7d6 a7a6 d6c7
info nodes 213051055 nps 1577419 time 135063
info currmove f8e7 currmovenumber 1
info depth 40 seldepth 0 multipv 1 score cp 28 lowerbound nodes 231165587 nps 15
77577 time 146532 pv f8e7 a7b8 e7d8 h6h5 d8d7 h5h4 h1f3 b8a7 d7c7 a7a6 f3b7 a6a7
 b7h1 a7a6 h1c6 a6a7 c6b7 h2h1n b7h1 a7a6 h1f3 a6a7 f3b7 h4h3 b7c8 h3h2 c8b7 h2h
1n b7h1 a7a6 c7d7 a6b6 d7e7 b6c7 h1f3 c7b6 e7f6
info nodes 231165587 nps 1577577 time 146532
info currmove f8e7 currmovenumber 1
info depth 40 seldepth 0 multipv 1 score cp 46 lowerbound nodes 234962167 nps 15
77912 time 148907 pv f8e7 a7b8 e7d8 b8a7 d8d7 a7b8 h1g2 b8a7 g2f3 a7b8 d7d8 h2h1
n f3h1 h6h5 d8d7 h5h4 d7d8 h4h3 d8d7 b8a7 d7c7 a7a6 h1b7 a6a7 b7c8 h3h2 c8b7 h2h
1n b7h1 a7a6 c7d7 a6b6 d7e7 b6c7 h1f3 c7b6 e7f6
info nodes 234962167 nps 1577753 time 148922
info currmove f8e7 currmovenumber 1
info depth 40 seldepth 0 multipv 1 score cp 73 lowerbound nodes 248871021 nps 15
74505 time 158063 pv f8e7 a7a6 e7d7 a6b6 h1d5 b6a7 d7c7 a7a6 d5h1 a6a7 h1b7 f5f4
 c7c8 f6f5 c8c7 f4f3 b7f3 a7a6 f3h1 a6a7 h1b7 f5f4 b7h1 a7a6 h1f3 a6a7 f3b7 f4f3
 b7f3 a7a6 f3b7 a6a7 b7h1 a7a6 h1f3 a6a7 f3b7 h6h5 b7h1 a7a6 h1f3 a6a7 f3b7 h5h4
 b7d5 a7a6 d5h1 a6a7 h1b7 h4h3 b7d5 a7a6 d5f3 a6a7 f3d5
info nodes 248871021 nps 1574355 time 158078
info currmove f8e7 currmovenumber 1
info depth 40 seldepth 0 multipv 1 score cp 114 lowerbound nodes 250651275 nps 1
571323 time 159516 pv f8e7 a7a6 e7d7 a6b6 h1d5 b6a7 d5f3 a7a6 d7c7 a6a7 f3c6 a7a
6 c6b7 a6a7 c7c8 a7b6 c8b8 h2h1n b7h1 h6h5 b8c8 h5h4 h1g2 b6a6 g2b7 a6b6 b7h1 h4
h3 h1b7 h3h2 b7h1 b6a7 c8c7 a7a6 h1b7 a6a7 b7d5 a7a6 d5h1 a6a7 h1c6 a7a6 c6b7
info nodes 250651275 nps 1571323 time 159516
info currmove f8e7 currmovenumber 1
info nodes 276291360 nps 1557667 time 177375
info currmove h1f3 currmovenumber 2
info currmove h1d5 currmovenumber 3
info currmove h1c6 currmovenumber 4
info currmove h1b7 currmovenumber 5
info currmove h1g2 currmovenumber 6
info currmove h1a8 currmovenumber 7
info currmove f8g7 currmovenumber 8
info currmove f8f7 currmovenumber 9
info currmove h1e4 currmovenumber 10
info currmove f8e8 currmovenumber 11
info currmove f8g8 currmovenumber 12
info depth 40 seldepth 58 multipv 1 score cp 161 nodes 276299486 nps 1557432 tim
e 177407 pv f8e7 a7a6 e7d7 a6b6 d7d6 h6h5 h1f3 h5h4 d6d7 h4h3 f3h1 f5f4 d7c8 f6f
5 c8b8 b6a6 h1f3 h2h1b f3h1 f4f3 h1f3 h3h2 b8c7 f5f4 c7b8 a6b6 f3b7 f4f3 b7f3 b6
a6 b8c7 a6a7 f3b7 h2h1n b7h1 a7a6 h1b7 a6a7 c7c6 a7b8 b7a6

[George Tsavdaris]

Sting-sf-3 (beta) gets this position, without ugly pawns h7 and f7 (only time consuming) on my dualcore in two minutes.
5K2/k7/5p1p/1p3p2/bP6/1p6/1P5p/7B w - - 0 1 bm Ke8(e7);

Pawn on f7 is critical in order to entice computer to capture it.
So if you hate by all means positions where f7 and h7 squares does not have any Pawns, then you should try this instead:
[d]8/k4K2/5p1p/1p3p2/bP6/1p6/1P5p/7B w - - 0 1


Also if you don't want so many Pawns you should try something like this:
[d]5K2/k4p1p/8/1p3p2/bP6/1p6/1P5p/7B w - - 0 1

But anyway the original position is the real one that programs should solve.
[d]5K2/k4p1p/5p1p/1p3p2/bP6/1p6/1P5p/7B w - - 0 1

I hope one day they will be able to solve it. And this is because if you see the very easy pattern, then it's a piece of cake for a computer to calculate it. But their skeleton is alpha beta and it is restricting them too much. Something has to be done about this, but of course it's not easy as it would have already been implemented.
Even worse is this: http://rybkaforum.net/cgi-bin/rybkaforu ... #pid465220

[d]8/8/8/7p/1p6/1p3p2/pP2p3/K1k1B3 w - - 0 21

It's unacceptable that the engines in 2013 have these evaluation for this position. But i guess the problem is that they have been designed for playing better Chess and the goal of their authors is to make them stronger and have not been designed for understanding Chess.

[peter]

[d]8/8/8/7p/1p6/1p3p2/pP2p3/K1k1B3 w - - 0 21

It's unacceptable that the engines in 2013 have these evaluation for this position.

Code: Select all

8/8/8/7p/1p6/1p3p2/pP2p3/K1k1B3 w - - 0 1

Analysis by Stockfish 2.3.1 JA 64bit SSE4.2:

...
21.Lf2 h4 22.Lxh4 Kd2 23.Lg5+ Kd1 24.Lh4 Kc2 25.Le1 Kd3 26.Lf2 Kc4 27.Lh4 Kc5 28.Lg3 Kc6 29.Lh4 Kb7 30.Lg3 Kc8 31.Lf2 Kc7 32.Lg3+ Kd7 33.Lh4 Kc6 34.Lf2 Kc7 
  =  (0.00)   Tiefe: 52/72   00:06:40  4232mN
21.Lf2 h4 22.Lxh4 Kd2 23.Lg5+ Kd1 24.Lh4 Kc2 25.Le1 Kd3 26.Lf2 Kc4 27.Le1 Kd5 28.Lh4 Kc4 29.Le1 
  =  (0.00)   Tiefe: 53/72   00:07:19  4618mN
21.Lf2 h4 22.Lxh4 Kd2 23.Lg5+ Kd1 24.Lh4 Kc2 25.Le1 Kd3 26.Lf2 Kc4 27.Le1 Kd5 28.Lh4 Kc4 29.Le1 
  =  (0.00)   Tiefe: 54/72   00:08:05  5074mN

What about the study's origin, George?

[George Tsavdaris]

[d]8/8/8/7p/1p6/1p3p2/pP2p3/K1k1B3 w - - 0 1

Analysis by Stockfish 2.3.1 JA 64bit SSE4.2:

21.Lf2 h4 22.Lxh4 Kd2 23.Lg5+ Kd1 24.Lh4 Kc2 25.Le1 Kd3 26.Lf2 Kc4 27.Le1 Kd5 28.Lh4 Kc4 29.Le1
= (0.00) Tiefe: 54/72 00:08:05 5074mN

What about the study's origin, George?

It's a result of the original study i gave with the 1...Rxh1 that (as i've said earlier) i don't know its origin:
[d]1k6/3p4/1B6/4Pp1p/1p5R/1p4p1/pP3n2/K6n w - - 0 1


I put it to show that even after all the key moves have been played, engines are still not able to understand that it is a draw! So to be able to find the Rxh1 move from the root position, i guess it is gonna take them many many years more....