Discussion of computer chess matches and engine tournaments.
Moderator: Ras
Rebel
Posts: 7515Joined: Thu Aug 18, 2011 12:04 pmFull name: Ed Schröder
Post by Rebel Sun May 16, 2021 3:25 pm
Komodo Dragon : Knight Odds Handicap Match
http://rebel13.nl/pgn4web-3.05/live-test.html Code: Select all
. Tournament : Gauntlet
. Time Control : 40 moves in 10 minutes repeating
. Games : 7 games each engine
. Opponents : elo range 2534-2699
1. Benjamin 2699
2. Gaviota 0.85.1 2698
3. Fruit 2.1 2695
4. Fridolin 3.10 2683
5. Velvet 1.1.0 2685
6. Arasan 14.1 2652
7. Supernova 2.3 2652
8. Galjoen 0.41.1 ~2589
9. Maverick 1.5 2537
10 Marvin 2.0.0 2534
. Software : Arena 3.0
. Hash table : Default setting engines
. Openings : dragon.epd
. rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - c0 "-wnb1";
. rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKB1R w KQkq - c0 "-wng1";
Komodo Dragon II will follow later.
90% of coding is debugging, the other 10% is writing bugs.
Rebel
Posts: 7515Joined: Thu Aug 18, 2011 12:04 pmFull name: Ed Schröder
Post by Rebel Sun May 16, 2021 6:40 pm
-Nc3 finished : 3.5 / 10-Nf3 running now.
90% of coding is debugging, the other 10% is writing bugs.
Rebel
Posts: 7515Joined: Thu Aug 18, 2011 12:04 pmFull name: Ed Schröder
Post by Rebel Sun May 16, 2021 10:07 pm
-Ng1 run : 4-6, 40%.
Again, a lot better than SF13, 0%.
PGN at -
http://rebel13.nl/dump/dragon-1-handicap.pgn
Will repeat with Dragon 2, thereafter SF13 with this engine pool.
90% of coding is debugging, the other 10% is writing bugs.
Rebel
Posts: 7515Joined: Thu Aug 18, 2011 12:04 pmFull name: Ed Schröder
Post by Rebel Mon May 17, 2021 9:37 am
-Nb1 : 3.5 - 6.5, 35% for Dragon II.
Code: Select all
Name Gms : Win : Draw : Lose : Pts : S-B
Komodo-dragon-2 10 : 3+ : 1= : 6- : 3.5 : 0.25
Arasan 14.1 1 : 1+ : 0= : 0- : 1.0 : 3.50
Benjamin 1 : 1+ : 0= : 0- : 1.0 : 3.50
Galjoen 0.41.1 1 : 1+ : 0= : 0- : 1.0 : 3.50
Gaviota-0.85.1 1 : 1+ : 0= : 0- : 1.0 : 3.50
Marvin 2.0.0 1 : 1+ : 0= : 0- : 1.0 : 3.50
Supernova 2.3 1 : 1+ : 0= : 0- : 1.0 : 3.50
Maverick 1.5 1 : 0+ : 1= : 0- : 0.5 : 1.75
Fridolin 3.10 1 : 0+ : 0= : 1- : 0.0 : 0.00
Fruit 2.1 1 : 0+ : 0= : 1- : 0.0 : 0.00
Velvet 1.1.0 1 : 0+ : 0= : 1- : 0.0 : 0.00
Games played 10, games to go 0
Tournament finished
-Ng1 now...
90% of coding is debugging, the other 10% is writing bugs.
lkaufman
Posts: 6284Joined: Sun Jan 10, 2010 6:15 amLocation: Maryland USAFull name: Larry Kaufman
Post by lkaufman Mon May 17, 2021 4:26 pm
Rebel wrote: ↑ Mon May 17, 2021 9:37 am -Nb1 : 3.5 - 6.5, 35% for Dragon II.
Code: Select all
Name Gms : Win : Draw : Lose : Pts : S-B
Komodo-dragon-2 10 : 3+ : 1= : 6- : 3.5 : 0.25
Arasan 14.1 1 : 1+ : 0= : 0- : 1.0 : 3.50
Benjamin 1 : 1+ : 0= : 0- : 1.0 : 3.50
Galjoen 0.41.1 1 : 1+ : 0= : 0- : 1.0 : 3.50
Gaviota-0.85.1 1 : 1+ : 0= : 0- : 1.0 : 3.50
Marvin 2.0.0 1 : 1+ : 0= : 0- : 1.0 : 3.50
Supernova 2.3 1 : 1+ : 0= : 0- : 1.0 : 3.50
Maverick 1.5 1 : 0+ : 1= : 0- : 0.5 : 1.75
Fridolin 3.10 1 : 0+ : 0= : 1- : 0.0 : 0.00
Fruit 2.1 1 : 0+ : 0= : 1- : 0.0 : 0.00
Velvet 1.1.0 1 : 0+ : 0= : 1- : 0.0 : 0.00
Games played 10, games to go 0
Tournament finished
-Ng1 now...
I finished my own knight odds blitz test, in which Dragon 2 and Stockfish 13 (each with Contempt 100) give knight odds (using the middle 200 positions from ChrisW opening set for the odds) at 2' + 1" to four opponents: Benjamin, Wasp 1.01, Fruit 2.2.1, and Naum 4. All engines one thread, all matches 200 games with the same openings. CCRL blitz ratings used for stats. Dragon 2 performed about a hundred elo better than SF 13 against the first three opponents, while SF 13 performed over a hundred elo better against Naum 4. I don't know the reason for this. Overall Dragon 2 performance rating was 2762, SF 13 was 2731. Of course you will get significantly lower performance ratings at your much longer time control since the odds-giver needs serious mistakes by the opponent to win or draw. Whether Dragon will outperform SF in your test, and if so by more or less than my 31 elo margin, will be interesting. Regarding the performance of the four engines receiving the handicap, Benjamin was outstanding, performing about a class (200 elo) better relative to its rating than Wasp and Fruit did, while Naum performed over a class worse relative to its rating than Wasp and Fruit did. Note that the average rating of Dragon 2 and Stockfish 13 is 3610, so this test gives a value of knight odds of 864 elo, but this is using the Bayes-Elo contracted ratings of CCRL. With normal Elo (Ordo), I think that the calculation would have come out right about my 1000 estimate for knight odds. I redid the math using CEGT ratings, which do use Ordo, and I got that knight odds = 984 elo!
Komodo rules!
Rebel
Posts: 7515Joined: Thu Aug 18, 2011 12:04 pmFull name: Ed Schröder
Post by Rebel Mon May 17, 2021 5:42 pm
- Ng1 : 2-8, 20%, making it 5.5 / 20.
PGN -
http://rebel13.nl/dump/dragon-2-handicap.pgn
Will run SF13 and thereafter Lc0
Current standing
world championship Knight odds Code: Select all
Komodo Dragon 1 7.5 points
Komodo Dragon 2 5.5 points
Stockfish 13 ..........
Lc0 .......... 90% of coding is debugging, the other 10% is writing bugs.
Rebel
Posts: 7515Joined: Thu Aug 18, 2011 12:04 pmFull name: Ed Schröder
Post by Rebel Tue May 18, 2021 8:00 am
Knight odds -Nb1 and Ng1
Code: Select all
Engine Points Games
1. Komodo Dragon 1 7.5 20
2. Stockfish 13 6.0 20
3. Komodo Dragon 2 5.5 20
4. Lc0 ... 20
Lc0 running now...
http://rebel13.nl/pgn4web-3.05/live-test.html 90% of coding is debugging, the other 10% is writing bugs.
Rebel
Posts: 7515Joined: Thu Aug 18, 2011 12:04 pmFull name: Ed Schröder
Post by Rebel Tue May 18, 2021 8:28 am
[pgn][Event "Lc0-handicap-nb1"]Laugh or cry ?
90% of coding is debugging, the other 10% is writing bugs.
Rebel
Posts: 7515Joined: Thu Aug 18, 2011 12:04 pmFull name: Ed Schröder
Post by Rebel Tue May 18, 2021 9:27 am
Given up on Lc0. While Arena dictates a -99.99 score for resignation the engine apparently resigns by itself.
90% of coding is debugging, the other 10% is writing bugs.
lkaufman
Posts: 6284Joined: Sun Jan 10, 2010 6:15 amLocation: Maryland USAFull name: Larry Kaufman
Post by lkaufman Tue May 18, 2021 4:11 pm
Rebel wrote: ↑ Tue May 18, 2021 9:27 am
Given up on Lc0. While Arena dictates a -99.99 score for resignation the engine apparently resigns by itself.
That's strange; I found that recent Lc0 (largest net) was quite strong giving knight odds, either to me or to engines. I didn't see stupid moves. But I don't use Arena, perhaps there is some hidden problem there?
Komodo rules!
