ShashChess

[Uri Blass]

The wait is over! Alexander 8.0 is officially out!

Take your chess training to the ultimate level with the most advanced "Glass Box" engine.

What’s new in version 8.0?
Aligned with Stockfish 18 – Unmatched raw playing strength!
Improved Handicap Mode – Train against a highly realistic, human-like opponent that adapts to your Elo and makes natural mistakes under pressure.
Major Bug Fixes – Smoother, faster, and more reliable than ever.

Stop analyzing with inscrutable "black boxes." Let Alexander explain the why behind every move and become your personal Expert Mentor!

Download Alexander 8.0 here:
https://github.com/amchess/Alexander/releases/tag/8.0

Discover all features in our new Wiki:
https://github.com/amchess/Alexander/wiki

Watch the Official Presentation:


Dive deep with our detailed YouTube Playlist:


Level up your game today!

There are many alexander files with no explanation how to use them and I think that you need some interface to have engine that explain things to you and only uci engine can only play chess

[Tibono]

There are many alexander files with no explanation how to use them and I think that you need some interface to have engine that explain things to you and only uci engine can only play chess

Hi,
I suspect you need a feature such as Arena's engine log window thru which you can input some commands adressed to the relevant engine. The use case would be: play against the engine and once you reach a position you need detailed analysis on, inject the 'eval' command that will dump a number of evaluation criteriae into the log. Either use it immediately, or copy-paste it to some notepad, play a candidate move, and repeat the command to get a refreshed list of evaluation criteriae. Then compare both output, looking for significant changes.
I understand it can be of some interest, yet cumbersome and requiring you perform "manually" a lot of comparison work. I would expect more advanced tools to automatize temporarily storing output and pointing out relevant changes thru a dedicated window display, managed by the software aside the engine itself. Hopefully a next step?

[amchess]

There are many alexander files with no explanation how to use them and I think that you need some interface to have engine that explain things to you and only uci engine can only play chess

Hi,
I suspect you need a feature such as Arena's engine log window thru which you can input some commands adressed to the relevant engine. The use case would be: play against the engine and once you reach a position you need detailed analysis on, inject the 'eval' command that will dump a number of evaluation criteriae into the log. Either use it immediately, or copy-paste it to some notepad, play a candidate move, and repeat the command to get a refreshed list of evaluation criteriae. Then compare both output, looking for significant changes.
I understand it can be of some interest, yet cumbersome and requiring you perform "manually" a lot of comparison work. I would expect more advanced tools to automatize temporarily storing output and pointing out relevant changes thru a dedicated window display, managed by the software aside the engine itself. Hopefully a next step?

Yes. We work on it: an uci engine can also explain the moves, not only play/suggest them...
Thanks.
Andrea

[amchess]

A new option Anti Dancing analysis:
https://github.com/amchess/ShashChess/releases/tag/41.1
https://github.com/amchess/Alexander/releases/tag/8.1

[criko]

A new option Anti Dancing analysis:
https://github.com/amchess/ShashChess/releases/tag/41.1
https://github.com/amchess/Alexander/releases/tag/8.1

Thx for the new version

[peter]

I fear, the Anti Dancing Analysis option is somewhat broken, at least with my avx512- compile I get rather weird output- eval- combinations at some of the anti engine puzzle- postions, I tried so far.
E.g. here's a well known Rudolph- study, White to move and draw, best move 1.Ba4+:

https://www.yacpdb.org/#search/M0I0XC8x ... MS8xLzA=/1

With ADA (Anti Dance Analysis) true and 30 threads of a 16x4.3GHz CPU:

[d]3B4/1r2p3/r2p1p2/bkp1P1p1/1p1P1PPp/p1P1K2P/PPB5/8 w - -

Engine: ShashChess41.1GD (8192 MB)

65/81 3:48 0.00 1.c4+ Kxc4 2.Ld3+ Kd5 3.Le4+ Ke6
4.Ke2 f5 5.Lxb7 fxg4 6.Lxa6 axb2
7.Lc4+ d5 8.La6 g3 9.Lc8+ Kf7 10.Lf5 b1D
11.Lxb1 Lxd8 12.fxg5 c4 13.Lf5 b3
14.axb3 (6.918.848.754) 30253

0.00? 1.c4(?) is clearly lost for White, as well as the rest of the output- line is, yet the engine gets stuck at 0.00 even at the end of this one given line.

With just a single thread and ADA false:

3B4/1r2p3/r2p1p2/bkp1P1p1/1p1P1PPp/p1P1K2P/PPB5/8 w - -

Engine: ShashChess41.1GD (1024 MB)


32/55 2:55 -6.05 1.La4+ Kxa4 2.b3+ Kb5 3.c4+ Kc6
4.d5+ Kd7 5.e6+ Kxd8 6.f5 Kc8 7.Ke4 Tb8
8.Ke3 Kc7 9.Kf3 Tg8 10.Kf2 Kd8
11.Kf3 Ke8 12.Kg2 Kf8 13.Kf3 Tg7
14.Kg2 (427.005.172) 2430
...

36/80 3:33 -3.91 1.La4+ Kxa4 2.b3+ Kb5 3.c4+ Kc6
4.d5+ Kd7 5.e6+ Kc8 6.f5 Kxd8 7.Ke2 Kc7
8.Kf2 Tba7 9.Kg2 Kb7 10.Kf1 Ta8
11.Ke1 Th8 12.Kf2 Tg8 13.Ke2 Taa8
14.Ke1 (579.217.133) 2718

Eval's still way to high, but it starts to decrease, with SMP of course faster, and anyhow the output- line shows the way to draw, aftetr 6.f5 the fortress is fixed.
That's only one of some examples I saw so far like this, wrong 0.00- eval (with MultiPV most of such examples show many to all output- lines having the same 0.00) not corresponding with the moves of the lines. In test suites of such anti engine puzzles, ADA true scores much less solved positions than ADA false does at same hardware- TC, regards

[peter]

Edit- time over, just wanted to show some more output to the given position with MultiPV=3 and ADA checked (true):

3B4/1r2p3/r2p1p2/bkp1P1p1/1p1P1PPp/p1P1K2P/PPB5/8 w - -

Engine: ShashChess41.1GD (8192 MB)

21 0:16 0.00 1.c4+ Kxc4 2.Ld3+ Kd5 3.Le4+ Ke6
4.Ke2 f5 5.Lxb7 axb2 6.Lxa6 fxg4
7.Lxa5 Kd5 8.Ld3 gxh3 9.exd6 exd6
10.Kf2 h2 (678.022.108) 41601

21 0:16 0.00 1.La4+ Kxa4 2.b3+ Kb5 3.c4+ Kc6
4.d5+ Kd7 5.e6+ Ke8 6.Lxa5 Tba7 7.f5 Txa5
8.Kf3 Kd8 9.Kf2 Ta4 10.Kf3 Ta8
11.Kf2 Tb8 (678.022.108) 41601

20 0:16 -3.73 1.exf6 d5 2.dxc5 gxf4+ 3.Kf3 Txf6
4.c6 Tc7 5.bxa3 b3 6.axb3 Kxc6 7.g5 Tf8
8.Lxc7 Kxc7 9.b4 e5 10.bxa5 e4+
11.Ke2 Tg8 12.a6 Txg5 13.c4 Tg2+
14.Kd1 (678.022.108) 41601


Again 30 threads, but this time MultiPV=3, Ba4 with correct output- line and eval (0.00) arises soon, yet 1.c4 still gets 0.00 too

[amchess]

I fear, the Anti Dancing Analysis option is somewhat broken, at least with my avx512- compile I get rather weird output- eval- combinations at some of the anti engine puzzle- postions, I tried so far.
E.g. here's a well known Rudolph- study, White to move and draw, best move 1.Ba4+:

https://www.yacpdb.org/#search/M0I0XC8x ... MS8xLzA=/1

With ADA (Anti Dance Analysis) true and 30 threads of a 16x4.3GHz CPU:

[d]3B4/1r2p3/r2p1p2/bkp1P1p1/1p1P1PPp/p1P1K2P/PPB5/8 w - -

Engine: ShashChess41.1GD (8192 MB)

65/81 3:48 0.00 1.c4+ Kxc4 2.Ld3+ Kd5 3.Le4+ Ke6
4.Ke2 f5 5.Lxb7 fxg4 6.Lxa6 axb2
7.Lc4+ d5 8.La6 g3 9.Lc8+ Kf7 10.Lf5 b1D
11.Lxb1 Lxd8 12.fxg5 c4 13.Lf5 b3
14.axb3 (6.918.848.754) 30253

0.00? 1.c4(?) is clearly lost for White, as well as the rest of the output- line is, yet the engine gets stuck at 0.00 even at the end of this one given line.

With just a single thread and ADA false:

3B4/1r2p3/r2p1p2/bkp1P1p1/1p1P1PPp/p1P1K2P/PPB5/8 w - -

Engine: ShashChess41.1GD (1024 MB)


32/55 2:55 -6.05 1.La4+ Kxa4 2.b3+ Kb5 3.c4+ Kc6
4.d5+ Kd7 5.e6+ Kxd8 6.f5 Kc8 7.Ke4 Tb8
8.Ke3 Kc7 9.Kf3 Tg8 10.Kf2 Kd8
11.Kf3 Ke8 12.Kg2 Kf8 13.Kf3 Tg7
14.Kg2 (427.005.172) 2430
...

36/80 3:33 -3.91 1.La4+ Kxa4 2.b3+ Kb5 3.c4+ Kc6
4.d5+ Kd7 5.e6+ Kc8 6.f5 Kxd8 7.Ke2 Kc7
8.Kf2 Tba7 9.Kg2 Kb7 10.Kf1 Ta8
11.Ke1 Th8 12.Kf2 Tg8 13.Ke2 Taa8
14.Ke1 (579.217.133) 2718

Eval's still way to high, but it starts to decrease, with SMP of course faster, and anyhow the output- line shows the way to draw, aftetr 6.f5 the fortress is fixed.
That's only one of some examples I saw so far like this, wrong 0.00- eval (with MultiPV most of such examples show many to all output- lines having the same 0.00) not corresponding with the moves of the lines. In test suites of such anti engine puzzles, ADA true scores much less solved positions than ADA false does at same hardware- TC, regards

Hi Peter, thank you for the detailed report and for testing the engine!
What you observed is exactly the expected side-effect of the 'Anti Dancing Analysis' option, and it's the reason why it defaults to 'false'.

To prevent the engine from repeating positions when it's winning, the ADA patch aggressively forces a hard 0.00 score (fail-low) as soon as it spots a repetition in the search tree. While this brilliantly cleans up the PV in standard winning positions (removing the annoying 'dancing' of pieces), it creates 'false draw mirages' in highly complex puzzles and fortresses. If a losing line accidentally contains a repetition deep in the search tree, the engine gets tricked into thinking it found a 0.00 draw, masking the actual loss.

ADA is a 'cosmetic/pedagogical' tool designed for human analysts who want a clean PV in standard converted games. For solving anti-engine puzzles, tactical suites, or regular play, ADA must absolutely remain OFF, as it alters the objective strength and the search tree of the engine. Thanks again for highlighting this!

[peter]

Hi Andrea!
Thanks for clarifying, regards