Stefan Pohl Computer Chess

private website for chessengine-tests


Anti Draw Openings - The future of Computerchess

 

 

Download of the Anti Draw Openings V2.1 collection here

 

This openings-sets collection contains 7 different openings concepts:


Chess324 unbalanced, Drawkiller V5, No Black Short Castling V2, No Black Castling V2, White One More Pawn V1, White One More Move V1 and Unbalanced Human Openings V3 (2021) & Unbalanced Human Openings 2022


Additionally, the download contains the tools, which are needed for the Armageddon- and my Advanced Armageddon rescoring of results. And my Gamepairs Rescorer Tool, for rescoring results like this:
"Thinking uniquely in game pairs makes sense with the biased openings used these days. While pentanomial makes  sense it is a bit complicated so we could simplify and score game pairs only (not games) as W-L-D (a traditional score of 2-0, or 1.5-0.5 is just a W)."

 

*** HOW TO USE *** Except the NBSC/NBC openings, all other openings concepts contain different openings-sets with increasing advantage for white, the evaluation-interval is part of the filename. This makes it possible to control the draw rate of the computerchess-events: I recommend to try the first or second "stage" of eval first, when using one of my AntiDraw-openings and look at the draw-rate of the results in your testsetup/engine-tournament. Because the draw-rate is highly influenced by thinking-time, PC-speed and strength of the playing engines. The draw-rate should be in an interval of 45%-60% (Drawkiller: 35%-50%) If the rate is lower, use a lower "stage" of evals (advantage for white). If the rate is  higher, switch to a higher "stage" of evals.


Explanation of the different openings concepts:

 

Chess324 64x / xxl: Chess324 is a subset of DFRC, with kings and rooks on their "normal" positions of classical chess (kings e1/e8, rooks a1,h1/a8,h8). So, in Chess324 no special chess-rules (castling-moves!) are needed. DFRC is FischerRandomChess (also known as Chess960) but including non-symmetrical positions of the pieces of white and black...

Chess324 64x means: All Chess324 positions combined with 2 single-step pawn-plies (example 1.a3 c6 or 1.e3 h6 and so on): 64 of these 2 single-step pawn-plies are possible, so there are 324 * 64 = 20736 possible different openings for evaluating.
The testing results show, that the classical Chess324 opening set has a pretty high draw-rate. My Chess324 64x sets give white a measuerable advantage, so the draw-rate is much lower. And all of my openings sets have at least 500 lines, which is enough for a 1000 game head-to-head competition (the classical Chess324 set contains 324 lines, only). But, compared to other openings concepts, the Chess324 64x openings sets are pretty small, because I had only 20736 opening lines as raw-data for evaluating, which is a very small number, compared to the other opening concepts... Additionally, I tried to build Chess324 openings with NBC/NBSC (no castlings for black, see below). But unfortunately, this did not work: The draw-rates were quite good, but the Elo-spreading of the results were bad...

Chess324 xxl means: All Chess324 positions combined with 6 single-step pawn-plies. Example: 1. h3 h6 2. g3 d6 3. d3 b6... So more than 600000 different Chess324 openings were possible and evaluated!!! The Chess324 xxl sets are around 30x bigger than the Chess324 64x sets...

 

Drawkiller: Main idea is, that the kings are on the opposite corner of the board at the beginning of the game (kings on a1 and h8 or kings on h1 and a8), with the queens not on the same center-line (queens on d1 and e8 or queens on e1 and d8). 

Drawkiller classic 2,3,4 moves (contains Drawkiller lines, followed by 2,3,4 pawn moves and Drawkiller QR-switch lines, followed by 2,3,4 pawn moves), And a bigger 4 moves file with a bigger eval interval of 0.29 instead of 0.09. Drawkiller human 4,5,6 moves: One of the 4 possible Drawkiller "endpositions" (2x classic and 2x QR-switch) as FEN-code followed by 4,5,6 human opening moves out of the Megabase 2020 (both players 2400+ Elo). Because the human openings are played out of the normal chess starting position, not out of one of the 4 Drawkiller positions, a lot of move-lines were illegal and had to be sorted out. And a bigger 6 moves file with a bigger eval interval of 0.29 instead of 0.09.

 

NBSC/NBC (=No Black Short Castling V2, No Black Castling V2): Openings, where black is not allowed to castle short (NBSC) or not allowed to castle long or short (NBC). That gives white a clear advantage, of course...

 

WOMM: White has one more move in the opening: In all of the endpositions, black has to move first (odd number of plies). These endpositions were FEN-coded and then the flag, defining the color, which has the first move, was changed  from "b" (=Black) to "w" (=White).

This means, that white had played the last move, leading to the endposition and is now allowed to make another move, before black has to move. That gives white a clear advantage, of course...

 

WOMP: White has one more pawn: Means, that an opening-line was created, in which the black pawn on a7 is deleted: 1. Na3 a5 2. Nb5 a4 3. Nc3 Nf6 4. Nxa4 Ng8 5. Nc3 Nf6 6. Nb1 Ng8 

Behind these 6 moves, human opening moves (4, 5 and 6 moves deep) out of the Megabase 2021 were added. That gives white a clear advantage, of course...

 

UHO: Unbalanced Human Openings: Openings with a measureable advantage for white in the endposition, but without any changes by me...
- 100% moves played by humans, only. Both players had at least 2300 Elo.
- opening lines played by the stronges players are at the beginning of the pgn/epd-files.
- no manually constructed openings (like my Drawkiller openings)
- no manually added moves to make castling impossible (like my NBSC openings)
- no selection of piece-patterns
UHO V3 (2021) contains opening-books, constructed out of the UHO big files. UHO 2022 doesnt. Because the community has shown no interest in theses opening books, I decided not to build them for UHO 2022. Because the effort is pretty high and the AntiDraw-Download package is already quite huge. But, as you can see in the testing results below, UHO V3 and UHO 2022 both work very well and the results are quite the same. So, use the UHO V3 books, if you want opening books, they work...


*** IMPORTANT ***  For UHO 2022 and Chess324, I used KomodoDragon 2.6 for evaluating, instead of the old version 1.0. KomodoDragon 2.6 shows evals, which are around +0.20 higher, than KomodoDragon 1.0 does. So, the Eval intervals in the new UHO 2022 are +0.20 higher: UHO 2022 starts with Eval [+1.10;+1.19] instead of Eval [+0.90;+0.99 (UHO V3). But as you can see below in the testing results, using UHO 2022 with Eval [+1.10;+1.19] gives nearly the same Elo-spreading and draw-rates, UHO V3 with Eval [+0.90;+0.99] gives...


I believe, this opening collection with 7 different opening concepts is giving the perfect solution for all purposes in computerchess: Unbalanced Human Openings are 100% human chess moves - nothing else. Drawkiller and Chess324 64x are 100% constructed and 50% constructed (Drawkiller human), for the best statistical results.
NBSC/NBC openings, WOMP and WOMM openings are in between, too: Half constructed and half human. 


All these openings are classical chess, no chess variant (like Chess960), because all lines start from the classical chess starting position or from a FEN-code, where all normal chess-rules (including the classical castlings-rules) are used.
So, all these openings can be used by all chessengines and chessGUIs on the planet, because no special, new chess-rules are needed like the castling-rules in Chess960 (Fischer Random Chess).

 

Here the testing results with long thinking-time and comparsion with 4 classical openings-sets. I think, the results speak for themselves...

Stockfish 15 avx2 vs. KomodoDragon 3.1 avx2, 3min+1sec, singlethread, no bases, no ponder,
AMD Ryzen 3900 12-core (24 threads) notebook. Cutechess-cli. 1000 games each testrun.
First 500 openings of each set are used, replayed with reversed colors.
No resign-rules. Cutechess-cli ends games with 5 Syzygy bases.

 

Stockfish 15 point of view:

Balsa_v2724 openings:        1000 (+53,=942,-5)    Score: 52.4%, Elo: +16,  Draws: 94.2%
Hert 500 openings:           1000 (+94,=895,-11)   Score: 54.1%, Elo: +30,  Draws: 89.5%
Stockfish 8moves_V3:         1000 (+99,=882,-19)   Score: 54.0%, Elo: +28,  Draws: 88.2%
Feobos v3:                   1000 (+146,=827,-27)  Score: 56.0%, Elo: +42,  Draws: 82.7%

For comparsion:
Average result of the 4 classical openings:        Score: 54.1%, Elo: +30,  Draws: 88.7%


Chess324 (for comparsion):    648 (+121,=508,-19)  Score: 57.9%, Elo: +56,  Draws: 78.4%


Chess324_64x_+100_+109:      1000 (+328,=577,-95)  Score: 61.6%, Elo: +84,  Draws: 57.7%
Chess324_64x_+110_+119:      1000 (+385,=482,-133) Score: 62.6%, Elo: +90,  Draws: 48.2%
Chess324_64x_+120_+130:      1000 (+409,=466,-125) Score: 64.2%, Elo: +102, Draws: 46.6%

 

Chess324_xxl_+100_+109:      1000 (+341,=579,-80)  Score: 63.0%, Elo: +94,  Draws: 57.9%
Chess324_xxl_+110_+119:      1000 (+375,=531,-94)  Score: 64.0%, Elo: +102, Draws: 53.1%
Chess324_xxl_+120_+129:      1000 (+408,=457,-135) Score: 63.6%, Elo: +98,  Draws: 45.7%


NBSC_V2_4mvs:                1000 (+373,=445,-182) Score: 59.5%, Elo: +68,  Draws: 44.5%
NBSC_V2_6mvs:                1000 (+352,=453,-195) Score: 57.9%, Elo: +56,  Draws: 45.3%

 

NBC_V2_4mvs:                 1000 (+414,=373,-213) Score: 60.0%, Elo: +72,  Draws: 37.3%
NBC_V2_6mvs:                 1000 (+399,=375,-226) Score: 58.6%, Elo: +62,  Draws: 37.5%

 

WOMM_V1_6mvs_+090_+099:      1000 (+344,=564,-92)  Score: 62.6%, Elo: +90,  Draws: 56.4%
WOMM_V1_8mvs_+090_+099:      1000 (+347,=557,-96)  Score: 62.5%, Elo: +90,  Draws: 55.7%
WOMM_V1_6mvs_+100_+109:      1000 (+403,=460,-137) Score: 63.3%, Elo: +96,  Draws: 46.0%
WOMM_V1_8mvs_+100_+109:      1000 (+393,=486,-121) Score: 63.6%, Elo: +98,  Draws: 48.6%
WOMM_V1_6mvs_+110_+119:      1000 (+439,=389,-172) Score: 63.4%, Elo: +96,  Draws: 38.9%
WOMM_V1_8mvs_+110_+119:      1000 (+429,=412,-159) Score: 63.5%, Elo: +98,  Draws: 41.2%

 

WOMP_V1_4mvs_+090_+099:      1000 (+398,=483,-119) Score: 64.0%, Elo: +100, Draws: 48.3%
WOMP_V1_6mvs_+090_+099:      1000 (+386,=501,-113) Score: 63.6%, Elo: +98,  Draws: 50.1%
WOMP_V1_4mvs_+100_+109:      1000 (+435,=389,-176) Score: 63.0%, Elo: +92,  Draws: 38.9%
WOMP_V1_6mvs_+100_+109:      1000 (+429,=448,-123) Score: 65.3%, Elo: +110, Draws: 44.8%
WOMP_V1_4mvs_+110_+119:      1000 (+458,=358,-184) Score: 63.7%, Elo: +98,  Draws: 35.8%
WOMP_V1_6mvs_+110_+119:      1000 (+441,=369,-190) Score: 62.5%, Elo: +90,  Draws: 36.9%

 

Drawkiller:
dk_+0.90_+0.99_classic_3mvs: 1000 (+416,=432,-152) Score: 63.2%, Elo: +94,  Draws: 43.2%
dk_+0.90_+0.99_human_5mvs:   1000 (+412,=443,-145) Score: 63.4%, Elo: +96,  Draws: 44.3%
dk_+1.00_+1.09_classic_3mvs: 1000 (+451,=344,-205) Score: 62.3%, Elo: +88,  Draws: 34.4%
dk_+1.00_+1.09_human_5mvs:   1000 (+426,=391,-183) Score: 62.1%, Elo: +86,  Draws: 39.1%
dk_+1.10_+1.19_classic_3mvs: 1000 (+451,=327,-222) Score: 61.5%, Elo: +82,  Draws: 32.7%
dk_+1.10_+1.19_human_5mvs:   1000 (+450,=333,-217) Score: 61.6%, Elo: +84,  Draws: 33.3%


UHO (V3= old version of 2021):
UHO_V3_6mvs_+090_+099:       1000 (+359,=560,-81)  Score: 63.9%, Elo: +100, Draws: 56.0%
UHO_V3_8mvs_+090_+099:       1000 (+333,=578,-89)  Score: 62.2%, Elo: +88,  Draws: 57.8%
UHO_V3_6mvs_+100_+109:       1000 (+391,=485,-124) Score: 63.4%, Elo: +96,  Draws: 48.5%
UHO_V3_8mvs_+100_+109:       1000 (+394,=499,-107) Score: 64.3%, Elo: +104, Draws: 49.9%
UHO_V3_6mvs_+110_+119:       1000 (+420,=419,-161) Score: 63.0%, Elo: +92,  Draws: 41.9%
UHO_V3_8mvs_+110_+119:       1000 (+409,=442,-149) Score: 63.0%, Elo: +94,  Draws: 44.2%

 

UHO_2022_6mvs_+110_+119:     1000 (+351,=565,-84)  Score: 63.4%, Elo: +96,  Draws: 56.5%
UHO_2022_8mvs_+110_+119:     1000 (+352,=553,-95)  Score: 62.9%, Elo: +92,  Draws: 55.3%
UHO_2022_6mvs_+120_+129:     1000 (+402,=480,-118) Score: 64.2%, Elo: +102, Draws: 48.0%
UHO_2022_8mvs_+120_+129:     1000 (+384,=496,-120) Score: 63.2%, Elo: +94,  Draws: 49.6%
UHO_2022_6mvs_+130_+139:     1000 (+415,=443,-142) Score: 63.6%, Elo: +98,  Draws: 44.3%
UHO_2022_8mvs_+130_+139:     1000 (+412,=408,-180) Score: 61.6%, Elo: +82,  Draws: 40.8%


 

Andreas Strangmueller (FGRL) did 3 interesting testruns on a very powerful machine,
using my Drawkiller V4 openings and a 4 moves classical openings-set for comparsion:

 

CPU:          Dual Intel Xeon E5-2673 v4, 40 cores (80 threads)
OS:           Windows 10 64-Bit
Tool:         Cutechess-Cli, no draw and resign rules
Engines:      Stockfish 13 x64 BMI2, default settings
Hash-Table:   2048 MB
Time control: 60 seconds + 0.6 seconds
Tablebases:   None
Ponder:       Off
Each opening: Repeated with changing colors


Stockfish 13 x64 BMI2 32 threads vs. Stockfish 13 x64 BMI2 16 threads, 1000 games

Score always Stockfish 13 with 32 Threads point of view

 

4 moves classical openings   : 1000 (+67/=923/-10),   52.9% +20 Elo, Draws 92.3%
Drawkiller_balanced_V4_2mvs  : 1000 (+125/=861/-14),  55.6% +39 Elo, Draws 86.1%
Drawkiller_unbalanced_V4_2mvs: 1000 (+333/=517/-150), 59.2% +64 Elo, Draws 51.7%

 

Advanced Armageddon rescoring of the Drawkiller unbalanced result:
Drawkiller_unbalanced_V4_2mvs: 1007 (+683,=  0,-324), 67.8% +131 Elo, Draws 0%

 

Conclusions on this testruns, done by an independent tester:
Mention, that the testsetup of playing SF 13 vs. another SF 13 with less speed on a pretty strong hardware is a very, very drawish setup: Using the classical openings with 4 moves,  the draw-rate is enormous: 92.3%. My Drawkiller balanced openings gave a really nice Elo-spreading: nearly doubled (+39 Elo) compared to the classical openings (+20 Elo), but the draw-rate was lower but still pretty high (86.1%).
But, in the final testrun, using my brand new concept of unbalanced Drawkiller openings, the result was just mind-blowing: A more than 3x bigger Elo-spreading (+64 Elo) compared to the classical openings (+20) Elo and a draw-rate of only 51.7% (!!!). Incredible! And much lower (=better), than I expected. This is really a breakthrough in a new dimension and will prevent the draw-death of high-end computerchess at least for the next decade.

Results published by Andreas Strangmueller on talkchess, look here


 

All work of the Pohl AntiDraw Openings done by Stefan Pohl (except the generation of the pawn moves added to the  Drawkiller lines: Done by Hauke Lutz)

 

(C) 2021 Stefan Pohl (SPCC)(www.sp-cc.de)