Stefan Pohl Computer Chessprivate website for chessengine-tests
The SALC opening-book and opening-positions
When using SALC-openings, the chance for attacks towards the opponent king is much higher than using normal opening-books. Because of this, computerchess using SALC openings, will bring more action and fun to watch (and a measureable lower number of draws), because the faster the computers get, the higher the quality of computerchess get and the higher the draw-rate in engine-engine-matches get...so the computerchess is in danger to die the "draw-death" in the near future. So, using SALC openings will give computerchess a future beyond playing only draws or using strange and incorrect gambit-openings for a lower draw-rate! If anybody doubt that the draw-death of computerchess will come, just take a look at Andreas Strangmuellers excellent website: http://www.fastgm.de/time-control4.html Here you can see, how much the draw-rate is rising, when the thinking-time (increasing PC-speed in the future is equivalent to longer thinking-time today, of course) is raised. From 49.2% draws with bullet-speed (Komodo 9.3 (singlecore, only (!)) in selfplay with 20''+0.2'' vs. 10''+0.1'') to 78.8% draws with long thinking-time (5120''+51.2'' vs. 2560''+25.6'') (3000 games each testrun (!))
What is new in SALC V5 (compared to older SALC V1 - V3) ? SALC V5 contains half-closed SALC-positions, only. And that leads to measureable lower draw-rates, compared to older SALC-versions - take a look at the testing-results below. So, it is strongly recommended to use the new SALC V5 openings from now, only and not the older SALC-versions !!!
These are the half-closed filters: 3) On d-line or e-line at least one white and one black pawn (=one of both center-lines closed)
The idea is, that in these half-closed positions, the probability of fast and many capturing-moves is much lower, so it should took more time (and moves) to reach drawish endgame-positions. So, the probability of an interesting and long midgame should get higher...
The SALC V5 (hc) book is 10 move (20 plies) deep and contains more than 200000 moves (around 2x bigger, than SALC V3 book!).
a) 5000 SALC V5 full-closed and 10 moves deep positions, filtered out of the 25000 opening-lines file. The testset contains only positions with an evaluation-interval of [-0.31,-0.53] and [+0.30, +0.53]
Full-closed means, following conditions had to be true (in the endposition of each opening line): 1-5 (SALC half-closed filters (see above)) and additional: 6) On d-line or e-line at least one white and one black pawn (=one of both center-lines full-closed) and on the other center-line at least one pawn (=other center-line at least half-closed).
Here the result of the Stockfish framework openings: Games Completed = 1000 of 1000 (Avg game length = 1036.164 sec) Settings = RR/256MB/300000ms+3000ms/M 450cp for 4 moves, D 120 moves/EPD:C:\LittleBlitzer3\34700_ok.epd(32000) 1. asmFish 170426 x64 603.0/1000 286-80-draws: 634 (L: m=0 t=0 i=0 a=80) (D: r=148 i=232 f=39 s=1 a=214) 2. Komodo 10.4 x64 397.0/1000 80-286-634 (L: m=0 t=2 i=0 a=284) (D: r=148 i=232 f=39 s=1 a=214)
Here the result of FEOBOS openings (v20 (official release version), contempt 5 and played in Gauntlet-mode, so the best 500 positions of FEOBOS are played and repeated with reversed colors (the positions in the FEOBOS.epd files are sorted by "quality" (whatever that means)(=highest possible contempt, should deliver lowest possible draw-rate, Feobos can do))) Games Completed = 1000 of 1000 (Avg game length = 1019.828 sec)
Here the result of the HERT openigs by Thomas Zipproth, which I use for my Stockfish-testruns
Here the result of the Noomen Gambit-Lines (I converted the 247 lines to EPD-positions and recognized, there was one double in it - deleted that double, so 246 positions were left) Games Completed = 492 of 492 (Avg game length = 984.853 sec)
Games Completed = 1000 of 1000 (Avg game length = 944.640 sec) Settings = RR/256MB/300000ms+3000ms/M 450cp for 4 moves, D 120 moves/EPD:C:\LittleBlitzer\SALC_V2_10moves.epd(10000) 1. asmFish 170426 x64 620.5/1000 351-110-draws: 539 (L: m=0 t=0 i=0 a=110) (D: r=149 i=231 f=38 s=0 a=121) 2. Komodo 10.4 x64 379.5/1000 110-351-539 (L: m=0 t=0 i=0 a=351) (D: r=149 i=231 f=38 s=0 a=121)
Here the result of the 100 opening-lines of the TCEC Superfinals Season 9+10 (by. J.Noomen) Games Completed = 200 of 200 (Avg game length = 986.465 sec)
Here the result of the new SALC V5 10 moves half-closed openings: Games Completed = 1000 of 1000 (Avg game length = 925.121 sec) Settings = RR/256MB/300000ms+3000ms/M 450cp for 4 moves, D 120 moves/EPD:C:\LittleBlitzer\SALC_V5_hc_10m.epd(25000) 1. asmFish 170426 x64 616.0/1000 370-138-draws: 492 (L: m=0 t=0 i=0 a=138)(D: r=133 i=200 f=29 s=0 a=130) 2. Komodo 10.4 x64 384.0/1000 138-370-492 (L: m=0 t=0 i=0 a=370)(D: r=133 i=200 f=29 s=0 a=130)
Here the result of the SALC V5 8 move full-closed 500 positions opening-set, which I recommend for serious testwork (played in Gauntlet-mode, so all 500 positions were used): Games Completed = 1000 of 1000 (Avg game length = 926.511 sec) Settings = Gauntlet/256MB/300000ms+3000ms/M 450cp for 4 moves, D 120 moves/EPD:C:\LittleBlitzer\SALC_V5_8m_500.epd(500) 1. asmFish 170426 x64 664.5/1000 426-97-draws: 477 (L: m=0 t=0 i=0 a=97)(D: r=141 i=192 f=27 s=2 a=115) 2. Komodo 10.4 x64 335.5/1000 97-426-477 (L: m=0 t=0 i=0 a=426)(D: r=141 i=192 f=27 s=2 a=115)
Results overview:
Framework standard openings: Score 60.3% – 39.7%, draws: 63.4% FEOBOS v20 contempt 5: Score 58.7% - 41.3%, draws: 64.1% HERT 500: Score: 60.6% - 39.4%, draws: 60.4% Noomen Gambit-Lines: Score 59.1% - 40.9%, draws: 59.3% Old SALC V2: Score: 62.1% - 37.9%, draws: 53.9% Noomen TCEC Superfinal (Season 9+10): Score: 62.5% - 37.5%, draws: 50.0% SALC V5 half-closed: Score 61.6% - 38.4%, draws: 49.2% SALC V5 full-closed 500 positions: Score 66.5% - 33.5%, draws: 47.7%
Download the games of the FEOBOS, Noomen openings and both SALC V5 testruns here
Take a look at the draw-rates !! Must I say anything more?!? Especially the full-closed 500-position opening set result is absolute amazing! Take a look, how wide the Elo-range of the result is (66.5% - 33.5%, instead of 60.3% - 39.7% with standard openings). And the number of draws was lowered -25% (634 to 477 draws in 1000 games, each testrun) compared to the standard Stockfish framework openings! The 100 Noomen TCEC Superfinal (Season 9+10) openinglines have a very good result, too! But mention, that 30 of these 100 positions are SALC-positions or positions, in which castlings into SALC-positions are quite likely in the first few moves to play by the engines. And 100 positions are a very, very small openings-set - it cannot be used for huge tournaments or for creating an openingbook.
Take a look at the average game-duration: Using the Stockfish standard openingset, the average game-duration was 1036 seconds. And using SALC V5 openings, the average game-duration was only 925 seconds. So, nearly 11% less PC-time was needed, using SALC, which means, in the same time, using SALC, around 11% more games can be played on a PC in the same time!!
And, finally, a lot of people said, that SALC-positions are only a small part of all possible chess-openings (thats true), and that using SALC-openings for engine-play will lead to distorted results – and that is not true. Here the proof:
Using the new HERT openings-set (by Thomas Zipproth) for my Stockfish-testing was a great opportunity to compare the gamebases played with HERT (contains positions selected from the most played variations in Engine and Human tournaments) and played with my SALC openings So, here the results. Both gamebases were played with 3'+1'', singlecore, 512 MB Hash. The only difference was the opening-set (HERT / SALC)... 2x 15000 games (!)
HERT:
1 Stockfish 170526 bmi2 : 3346 7 7 5000 71.3 % 3171 45.6 %
Elo-differences: 1-2: 32
average game length: +13.7% compared to SALC games (moves), +10% compared to SALC games (time)
White Wins: 5129 (34.2 %)
SALC V3:
1 Stockfish 170526 bmi2 : 3359 7 7 5000 72.7 % 3168 39.9 %
Elo-differences: 1-2: 32
White Wins: 5476 (36.5 %)
Conclusions: 1) SALC lowers the draw-rate a lot (35.8%) , compared to the HERT openings-set (42.8%) - mention, that the HERT-set was optimized for a low draw-rate. Thomas Zipproth has chosen only lines, which were not too drawish. Using other "classical" openings-sets should lead to a higher draw-rate, than using HERT. 2) The order of rank is the same for all engines in both gamebases = no distorted results playing SALC.
|