Release : Chess System Tal EXTREME

[peter]

How to make book.bin get used?

Forget about that, with book.bin in same directory as binary it does get used, I just forgot to set not only Bookdepth but also Bookrandom >0.

With UHO 2024 900-990cp 6mvs., 3'+1" single thread (4.3GHz) I had two little matches, fist one against Dragon1, the earliest version of Dragon, that can be downloaded for free already, setting Personality to Aggressive (therefore Dragon1Aggr), which means playing with HCE instead of NNUE, costing much Elo of course, some more than e.g. MCTS costs. compared to default, second one against Patricia5:

Score of ChessSystemTalExtreme vs Dragon1Aggr: 180 - 142 - 178 [0.538]
Elo difference: 26.5 +/- 24.5, LOS: 98.3 %, DrawRatio: 35.6 %
500 of 500 games finished.

Score of Patricia5 vs ChessSystemTalExtreme: 189 - 10 - 109 [0.791]
Elo difference: 230.8 +/- 32.7, LOS: 100.0 %, DrawRatio: 35.4 %
308 of 500 games finished.

Broke this second one match off because of the too big difference to get >50 wins for CSTE and thus enough for a reasonable EAS- rating out of 500 games at least.
For an indirect comparison, I took this one match of Patricia against latest Arasan:

Score of Patricia5 vs Arasan25.3: 51 - 209 - 240 [0.342]
Elo difference: -113.7 +/- 21.9, LOS: 0.0 %, DrawRatio: 48.0 %
500 of 500 games finished.

Out of he CSTE- Dragon1A- match an EAS-Score for CSTE:

Code: Select all

                         early           bad  avg.win 
Rank  EAS-Score  sacs    sacs   shorts  draws  moves  Engine/player 
---------------------------------------------------------------------------
   1    185993  20.56%  10.81%  11.67%  11.80%   74   ChessSystemTalExtreme  
   2    103976  05.63%  25.00%  23.94%  38.20%   68   Dragon1Aggr  
-------------------------------------------------------------------
*** Average length of all won games:             71 moves
*** Movelimit for early sac bonus  :             35 moves

And out of the Arasan- Patricia- match:

Code: Select all

                         early           bad  avg.win 
Rank  EAS-Score  sacs    sacs   shorts  draws  moves  Engine/player 
---------------------------------------------------------------------------
   1    417177  43.14%  31.82%  31.37%  05.83%   68   Patricia5  
   2    130182  03.83%  50.00%  14.35%  50.42%   77   Arasan25.3  
-------------------------------------------------------------------
*** Average length of all won games:             75 moves
*** Movelimit for early sac bonus  :             37 moves

With the 256 positions at 8 threads and 30"/pos.:

Code: Select all

    Program                                    Elo   +/-  Matches  Score   Av.Op.   S.Pos.   MST1    MST2   RIndex
 36 Patricia5.0-8t-MuPV4                     : 3444    4   8348    41.6 %   3503   157/256    8.2s   16.6s   0.24
 37 ChessSystemTalExtreme-8t                 : 3418    4   7908    37.8 %   3505   114/256    5.2s   19.0s   0.35
 
MST1  : Mean solution time (solved positions only)
MST2  : Mean solution time (solved and unsolved positions)
RIndex: Score according to solution time ranking for each position

[Rebel]

Testrun for my full UHO ratinglist started. Result in 4-5 days. Will CS Tal Extreme be able to make the first place in my EAS-ratinglist?

I already do know the answer from my testing and the answer is : it depends how you use your tool.

This is when using the full version, Chess-System-Tal-Extreme tops, Patricia-5-xtremeEAS 70.000 EAS points less.

Code: Select all

                         early           bad  avg.win 
Rank  EAS-Score  sacs    sacs   shorts  draws  moves  Engine/player 
---------------------------------------------------------------------------
   1    773968  52.00%  21.44%  56.41%  05.60%   48   Chess-System-Tal-Extreme  
   2    738967  48.36%  20.68%  55.78%  06.40%   48   Rebel-Extreme-1.1  
   3    699636  56.72%  22.42%  47.34%  04.26%   52   Patricia-5-xtremeEAS  

Then use gauntlet version and then we get :

Code: Select all

                         early           bad  avg.win 
Rank  EAS-Score  sacs    sacs   shorts  draws  moves  Engine/player 
---------------------------------------------------------------------------
   1    782396  56.72%  28.16%  52.82%  04.26%   52   Patricia-5-xtremeEAS  
   1    773968  52.00%  21.44%  56.41%  05.60%   48   Chess-System-Tal-Extreme  
   1    713389  48.36%  20.68%  53.63%  06.40%   48   Rebel-Extreme-1.1  

And now suddenly Patricia-5-xtremeEAS has 80.00 EAS points more ??

Suddenly the numbers and percentages of Patricia and Rebel-Extreme are different ??

We recently has discussed this privately in email, you were unwilling to fix this.

And as I told you, the only good way is the gauntlet way, run each match PGN separately, each engine should be unique and not influenced by other engines.

Meaning that Patricia-5-xtremeEAS should top your list.

Check yourself, download the first 5 engines of the UHO playing style rating list.

[Rebel]

Brilliant! Thanks Ed and Chris.

Minor point, it reports nodes-per-second incorrectly:

Code: Select all

info depth 6 seldepth 5 time 7 nodes 330 score cp 17 pv e2e4 c7c5 b1c3 b8c6 hashfull 0 Knps 47142

Note the "Knps", this should be "nps".

— Steve

How attentive of you

[peter]

Thanks for the new version!
How to make book.bin get used?
Tried to put it in same directory as binary and to put it into a subdirectory I named polyglot-books, because that was the way it worked with older versions, regards

Should work.

Yet book isn't used for more than 5 moves at certain lines

At position after e.g. 1. d4 Nf6 2. c4 e6 3. Nf3 d5 4. Nc3 c6 5. Bg5 h6:

setoption name Bookfile value C:\eng\polyglot-books\book.bin
info string book set to C:\eng\polyglot-books\book.bin
setoption name Bookdepth value 200
info string set Bookdepth to 200
position fen rnbqkb1r/pp3pp1/2p1pn1p/3p2B1/2PP4/2N2N2/PP2PPPP/R2QKB1R w KQkq - 0 6
go infinite
bestmove g5h4

But with one more ply 6.Bh4:

position fen rnbqkb1r/pp3pp1/2p1pn1p/3p4/2PP3B/2N2N2/PP2PPPP/R2QKB1R b KQkq - 0 6
go infinite
info depth 1 seldepth 1 time 6 nodes 2 score cp -16 pv d5c4 hashfull 0 Knps 333
info depth 1 seldepth 1 time 6 nodes 11 score cp 3 pv g7g5 hashfull 0 Knps 1833
info depth 1 seldepth 3 time 6 nodes 42 score cp 6 pv f8e7 hashfull 0 Knps 7000
info depth 2 seldepth 3 time 6 nodes 61 score cp 6 pv f8e7 hashfull 0 Knps 10166
info depth 3 seldepth 3 time 6 nodes 227 score cp 6 pv f8e7 hashfull 0 Knps 37833
...

Engine starts calculating instead of playing 6...dxc4 out of book instantly, even if it's the move stored there with 100% probability. That's not so with each and every 5- moves long line, but e.g. Dragon uses the same book without problems at the given line as long as it's stored. BTW it's almost the same as Cerebellum3Merge.bin, that earlier versions of Rebel and CST came along with under name book.bin way back then, at the given line I just added 6.Bh4 with SCID and made 6.Bf6 not to be played actively (0%), even tried orginal Cerebellum3Merge.bin too, it's the same, even if there only 6.Bxf6 is to be played actively, book is used by engine as well yet 6th move isn't played out of book, regards

[pohl4711]

I get crashes in the Fritz 20 GUI. This doesn't happen with any other engine.

"Chess System Tal Extreme caused an exception. This is a problem in the engine"

Maybe that's an old Fritz issue, a too long executable name, rename it for instance to : CSTal-extreme.exe

??? I wrote, that up to 10 Threads, the engine runs in Fritz on my 12 core machine. Only using 11 or 12 leads to the crash.

[pohl4711]

Testrun for my full UHO ratinglist started. Result in 4-5 days. Will CS Tal Extreme be able to make the first place in my EAS-ratinglist?

I already do know the answer from my testing and the answer is : it depends how you use your tool.

I use my EAS-Tool, as it should be used: Running it on my full ratinglist database. When the testrun is finished.
And I think, the result of CS Tal Extreme will surprise you. After 4500 games (sadly already 2 disconnects by CS Tal Extreme - not good, but still acceptable), the EAS-Score looks very promising, when setting my Gauntlet-EAS-Tool to 77 moves, as the full ratinglist has as average value (thats how to do it). CS Tal Extreme will be clearly at rank #1 right now, if the results do not change drastically until the end of the testrun.
Results in 2-3 days.

[Peter Berger]

Yet book isn't used for more than 5 moves at certain lines

At position after e.g. 1. d4 Nf6 2. c4 e6 3. Nf3 d5 4. Nc3 c6 5. Bg5 h6:

I can explain this behaviour, as it is known from previous Rebel versions. 6. Bxf6 is most likely in your book, but the Rebel/Tal engines won't play moves from the Polyglot book that take material but rather think for themselves. If you reply with 6...Qxf6 the engine will most likely play from book again.

[Rebel]

Yet book isn't used for more than 5 moves at certain lines

At position after e.g. 1. d4 Nf6 2. c4 e6 3. Nf3 d5 4. Nc3 c6 5. Bg5 h6:

I can explain this behaviour, as it is known from previous Rebel versions. 6. Bxf6 is most likely in your book, but the Rebel/Tal engines won't play moves from the Polyglot book that take material but rather think for themselves. If you reply with 6...Qxf6 the engine will most likely play from book again.

Correct Peter, there is problem with book captures. Not sure when that slipped in.

[peter]

Yet book isn't used for more than 5 moves at certain lines

At position after e.g. 1. d4 Nf6 2. c4 e6 3. Nf3 d5 4. Nc3 c6 5. Bg5 h6:

I can explain this behaviour, as it is known from previous Rebel versions. 6. Bxf6 is most likely in your book, but the Rebel/Tal engines won't play moves from the Polyglot book that take material but rather think for themselves. If you reply with 6...Qxf6 the engine will most likely play from book again.

Correct Peter, there is problem with book captures. Not sure when that slipped in.

Ok, thanks for corfirmation, also to Peter (from)

[Peter Berger]

Quite a long technical phase at the end of the game, that you can safely ignore.
But moves 10-20 were pretty amazing. 18....Bxe5 probably simply wins the game ( I haven't checked with chess engines).
If Chess System Tal extreme continues playing this way, I will enjoy watching its games very much.

[pgn]
[Event "Lang 90min+30sek"]
[Site "Berlin"]
[Date "2026.01.09"]
[Round "?"]
[White "Crafty 25.6"]
[Black "Chess System Tal Extreme"]
[Result "0-1"]
[ECO "D45"]
[PlyCount "142"]
[TimeControl "5400+30"]

{4096MB, DESKTOP-8OCGGEO} 1. d4 {[%emt 0:00:00]} d5 {[%eval 31,29] [%emt 0:04:45]} 2. c4 {[%emt 0:00:07]} e6 {[%eval 38,26] [%emt 0:01:12]} 3. Nc3 {[%emt 0:00:07] (Sf3)} c6 {[%eval 70,26] [%emt 0:00:56]} 4. e3 {[%emt 0:00:06] (e4)} Nf6 {[%eval 21,26] [%emt 0:02:22]} 5. Nf3 {[%emt 0:00:06]} Nbd7 {[%eval 20,27] [%emt 0:03:49]} 6. Qc2 {[%emt 0:00:09] (Ld3)} Bd6 {[%eval 23,27] [%emt 0:03:05]} 7. g4 {[%emt 0:00:08] (Ld3)} h6 {[%eval 3,22] [%emt 0:00:45]} 8. h3 {[%emt 0:00:07] (Ld2)} e5 {[%eval 32,25] [%emt 0:02:21]} 9. Bd2 {[%emt 0:00:08]} O-O {[%eval 28,24] [%emt 0:00:44]} 10. cxd5 {[%emt 0:01:30] (0-0-0)} e4 {[%eval -24,29] [%emt 0:06:10]} 11. dxc6 {[%emt 0:03:31] (Sxe4)} exf3 {[%eval -24,27] [%emt 0:02:28]} 12. cxd7 {[%emt 0:00:43]} Bxd7 {[%eval -24,30] [%emt 0:03:30]} 13. e4 {[%emt 0:00:07]} b5 {[%eval -59,26] [%emt 0:02:07]} 14. e5 {[%emt 0:02:12]} Rc8 {[%eval -56,29] [%emt 0:02:43]} 15. a3 {[%emt 0:17:11] (g5)} Re8 {[%eval -29,26] [%emt 0:03:25]} 16. Bd3 {[%emt 0:12:50]} a5 {[%eval -79,29] [%emt 0:03:07]} 17. Qb1 {[%emt 0:01:29] (Kf1)} b4 {[%eval -157,26] [%emt 0:01:39]} 18. Nd1 {[%emt 0:01:36] (axb4)} Bxe5 {[%eval -220,27] [%emt 0:01:39]} 19. dxe5 {[%emt 0:01:42]} Rxe5+ {[%eval -231,27] [%emt 0:02:46]} 20. Ne3 {[%emt 0:01:16]} Nd5 {[%eval -233,30] [%emt 0:02:04]} 21. axb4 {[%emt 0:04:34] (Kf1)} Qh4 {[%eval -338,26] [%emt 0:03:59]} 22. Rxa5 {[%emt 0:00:10] (Ta3)} Rxe3+ {[%eval -319,27] [%emt 0:01:40]} 23. Bxe3 {[%emt 0:00:07]} Nxe3 {[%eval -339,30] [%emt 0:02:31]} 24. Rh5 {[%emt 0:01:17] (Le4)} Qe7 {[%eval -406,27] [%emt 0:01:27]} 25. Bh7+ {[%emt 0:05:12] (Tc5)} Kh8 {[%eval -434,27] [%emt 0:01:36]} 26. Qe4 {[%emt 0:00:38]} Ng2+ {[%eval -442,28] [%emt 0:01:06]} 27. Kd2 {[%emt 0:00:25]} Qf6 {[%eval -442,29] [%emt 0:01:28]} 28. Rc1 {[%emt 0:00:08] (Tb1)} Re8 {[%eval -512,27] [%emt 0:01:32]} 29. Rf5 {[%emt 0:00:20]} Qxb2+ {[%eval -497,28] [%emt 0:01:37]} 30. Rc2 {[%emt 0:01:19]} Qa1 {[%eval -506,32] [%emt 0:02:01]} 31. Ra5 {[%emt 0:00:20]} Qf1 {[%eval -524,32] [%emt 0:02:35]} 32. Re5 {[%emt 0:02:06]} Rxe5 {[%eval -550,31] [%emt 0:04:29]} 33. Qxe5 {[%emt 0:00:06]} Kxh7 {[%eval -572,29] [%emt 0:01:02]} 34. Kc3 {[%emt 0:01:38]} g6 {[%eval -573,31] [%emt 0:01:21]} 35. Qd5 {[%emt 0:00:08] (h4)} Qa1+ {[%eval -591,32] [%emt 0:01:10]} 36. Rb2 {[%emt 0:00:09]} Qc1+ {[%eval -609,34] [%emt 0:03:10]} 37. Rc2 {[%emt 0:00:08]} Qa3+ {[%eval -621,30] [%emt 0:00:45]} 38. Kd2 {[%emt 0:02:44] (Kd4)} Qxb4+ {[%eval -643,31] [%emt 0:01:25]} 39. Kc1 {[%emt 0:00:07]} Be6 {[%eval -660,31] [%emt 0:01:30]} 40. Qxf3 {[%emt 0:00:09]} Nf4 {[%eval -666,31] [%emt 0:00:44]} 41. Rd2 {[%emt 0:01:18]} Bb3 {[%eval -665,32] [%emt 0:02:02]} 42. Qg3 {[%emt 0:00:08] (Dc6)} Bd5 {[%eval -883,29] [%emt 0:00:51]} 43. Rc2 {[%emt 0:00:35]} g5 {[%eval -900,30] [%emt 0:00:56]} 44. Qe3 {[%emt 0:00:36]} Be4 {[%eval -1085,30] [%emt 0:01:03]} 45. Rc7 {[%emt 0:00:20]} Qb1+ {[%eval -1250,29] [%emt 0:00:46]} 46. Kd2 {[%emt 0:00:08]} Qa2+ {[%eval -1269,30] [%emt 0:00:57]} 47. Kc1 {[%emt 0:00:38]} Bg6 {[%eval -1328,30] [%emt 0:01:13]} 48. Kd1 {[%emt 0:01:30] (f3)} Bd3 {[%eval -1544,28] [%emt 0:00:54]} 49. Rc3 {[%emt 0:00:08]} Qa1+ {[%eval -2179,30] [%emt 0:01:02]} 50. Qc1 {[%emt 0:00:16]} Qa4+ {[%eval -32707,34] [%emt 0:01:28]} 51. Ke1 {[%emt 0:00:06]} Qe4+ {[%eval -32719,36] [%emt 0:00:39]} 52. Qe3 {[%emt 0:00:38]} Ng2+ {[%eval -32725,41] [%emt 0:00:06]} 53. Kd2 {[%emt 0:00:10]} Nxe3 {[%eval -32727,36] [%emt 0:00:28]} 54. Rxd3 {[%emt 0:01:45]} Nd5 {[%eval -32731,46] [%emt 0:00:04]} 55. Rb3 {[%emt 0:00:29]} Qd4+ {[%eval -32733,45] [%emt 0:00:05]} 56. Kc1 {[%emt 0:00:11] (Kc2)} Qxf2 {[%eval -32737,40] [%emt 0:00:14]} 57. Rb2 {[%emt 0:00:24]} Qe1+ {[%eval -32739,30] [%emt 0:00:23]} 58. Kc2 {[%emt 0:00:09]} Qc3+ {[%eval -32741,64] [%emt 0:00:26]} 59. Kb1 {[%emt 0:00:08]} Nb4 {[%eval -32743,68] [%emt 0:00:49]} 60. Rxb4 {[%emt 0:00:10]} Qxb4+ {[%eval -32745,69] [%emt 0:00:43]} 61. Kc2 {[%emt 0:00:07] (Kc1)} Qa3 {[%eval -32747,67] [%emt 0:00:42]} 62. Kd2 {[%emt 0:00:07]} Qxh3 {[%eval -32749,73] [%emt 0:00:53]} 63. Ke2 {[%emt 0:00:09] (Kc2)} Qxg4+ {[%eval -32751,77] [%emt 0:00:37]} 64. Kd3 {[%emt 0:00:08]} Qf4 {[%eval -32753,87] [%emt 0:00:46]} 65. Kc3 {[%emt 0:00:12] (Ke2)} g4 {[%eval -32753,52] [%emt 0:00:42]} 66. Kc2 {[%emt 0:00:07]} Qe3 {[%eval -32755,59] [%emt 0:00:38]} 67. Kb2 {[%emt 0:00:19]} g3 {[%eval -32757,116] [%emt 0:00:01]} 68. Ka2 {[%emt 0:00:10]} g2 {[%eval -32759,116] [%emt 0:00:01]} 69. Kb2 {[%emt 0:00:05]} g1=Q {[%eval -32761,116] [%emt 0:00:00]} 70. Ka2 {[%emt 0:00:08]} Qh2+ {[%eval -32763,116] [%emt 0:00:01]} 71. Ka1 {[%emt 0:00:06]} Qeg1# {[%eval -32765,116] [%emt 0:00:00]} 0-1
[/pgn]