Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Test Positions / Fortress positions Stockfish vs LeelaZero
- - By ChrisW (**) Date 2018-05-30 20:47
I collected a few fortress EPDs, not many, it doesn't seem to be a heavily researched topic, these below form the first part of a set I have marked "fortress.epd" where most of the set are not fortresses at all (imo).

I was expecting Leela to outperform Stockfish, and, on this very limited set, with very lax time control on my part, and not exactly optimal conditions, both programs did rather badly.

However, in the time allowed, Stockfish didn't get any at all, while Leela got three. Evaluations were quite interesting though.

2k5/2p5/1q1p4/pPpPp1p1/N1P1Pp1p/P4PbP/KQ4P1/8 w - -
; am Nxb6 + ? ; c1 diff=10;
 
2k5/2p5/1q1p4/pPpPp1pp/N1P1Pp2/P4PbP/KQ4P1/8 w - -
; am Nxb6 + ? ; c1 diff=10;
 
8/4nk2/1p3p2/2rp2pp/1P1R1N1P/6P1/3KPP2/8 b - -
; am Rb5 ? ; c1 diff=7;
 
5rkq/3r1p1p/5RpP/2p3P1/2B2Q2/p7/P4PK1/8 w - -
0 1; draw;
 
k1b5/1p1p1p1p/pPpPpPpP/P1P1P1P1/8/8/8/K1B5 w - -
0 1; draw;
 
k7/1p1p1p1p/pPpPpPpP/P1P1P1P1/1B1B1B1B/B1B1B1B1/1B1B1B1B/K1B1B1B1 w - -
0 1; draw;
 
k7/1p1p1p1p/pPpPpPpP/B1B1B1B1/1BRBRBRB/B1BQBQB1/KBRBRBRB/B1B1B1B1 w - -
0 1; draw;
 
k7/1p1p1p1p/pPpPpPpP/P1P1P1P1/2RRRR2/BRBRBRBR/KBRBRBRB/B1BRBRBR w - -
0 1; draw;
 
1b6/2p5/1kB5/1PN5/8/3K4/8/q7 w - -
; bm Nb3;;
 
8/1p2b3/6k1/5p2/p2NbP1p/P1B1P1p1/1P5P/6K1 w - -
; bm h3; c0 fortress From Smirin-HIARCS;
 
8/1p6/1p6/kPp2P1K/2P5/N1Pp4/q2P4/1N6 w - -
; draw; c0 eval = draw; c1 diff=9;
 
8/8/5k2/8/2B4p/5RbP/5pP1/7K b - -
; draw; c0 eval = draw; c1 diff=7;
 
8/8/8/5Bp1/7k/8/4pPKP/8 w - -
; bm Bg4;  eval = draw;;
 
r1b1k3/1p1p1pp1/1p1P3p/pP6/P7/8/5PPP/4R1K1 b q -
; c0 eval +/-; c1 diff=6;
 
1k3b1q/pP2p1p1/P1K1P1Pp/7P/2B5/8/8/8 w - -
; bm Kd5; Bd5; c0 eval +/-; c1 diff=8;
Parent - - By Antares (****) [de] Date 2018-05-30 20:51
k7/1p1p1p1p/pPpPpPpP/P1P1P1P1/1B1B1B1B/B1B1B1B1/1B1B1B1B/K1B1B1B1 w - -


Could you please provide us a PGN how you reached this position? :wink:

Yes, chess-engines are there to evaluate chess-positions...
Parent - - By Peter Grayson (****) [us] Date 2018-05-30 22:37
I fully agree. Some of those positions are not chess positions so it is ridiculous to expect an engine to find a solution to a nonsense position. Wouldn't be surprised if some engines and perhaps GUI's crash on being presented with these.

PeterG
Parent - - By ChrisW (**) Date 2018-05-31 09:29
Stockfish does indeed crash on the "too much" material one, I thought I'ld deleted it, but obvs not
Parent - - By Peter Grayson (****) [us] Date 2018-05-31 12:18
I do not criticise creating certain types of positions to test engines for their handling of those aspects of the game even if the positions may not have been produced in a game, indeed they can be interesting provided they conform to the rule they could be reached in a game. That ensures their "game validity".

PeterG
Parent - - By Chris Whittington (**) [fr] Date 2018-05-31 14:55
But you are on another waveband, comrade, doing something else. Why would I care for your concept game validity when in my context it's irrelevant?
Parent - - By Peter Grayson (****) [us] Date 2018-05-31 15:40

> Why would I care for your concept game validity when in my context it's irrelevant?


Haven't lost your barbed tongue approach I see!
Parent - By Chris Whittington (**) [fr] Date 2018-05-31 15:56
oleaginous chit chat reserved special cases only, comrade. raw truth on computer chess forums. Try Facebook for a bit more social.
Parent - - By ChrisW (**) Date 2018-05-31 09:27
it's an artificial fortress, designed to see if an engine can detect "no progress possible". I used them to develop working code to detect total blockages (which works btw).

but feel free to miss the point completely. we are not all "testers", you know, some of us "do" and some of us "work out how things work".
Parent - - By Antares (****) [de] Date 2018-05-31 17:14
I understand your test-aim, still you can achieve this with a legal chess-position and don't need 14 dark-squared bishops too much for that aim... i mean, why in hell should a NN like Leela bother with 14 dark-squared bishops too much?
Parent - - By ChrisW (**) Date 2018-05-31 17:27
no idea, I am not responsible for the creation of either the test position or the fortress set of test positions.

it's actually not at all fair to give that position to Leela because she will never have and never will see anything like it. Whereas, it is fair to give it to Stockfish, cos he just adds everything up, so it is not really a useful comparator position.

Why do ask negative questions? You could have asked which positions they differed in and made some progress in understanding. But you chose negative. In my psychological pack of tricks, this suggests you are more interested in being right than in truth. With result pointless back and forth. Is that fair comment?
Parent - - By Antares (****) [de] Date 2018-05-31 18:10
No no, i'm just in love with irony/sarcasm, but next time i will formulate it more truth-related. :smile:
Parent - By ChrisW (**) Date 2018-05-31 18:40
hey progress
Up Topic The Rybka Lounge / Test Positions / Fortress positions Stockfish vs LeelaZero

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill