I was going to say I think you went too far, it would be tough for Dragon to approach 50%, but as I write this it is dead even!
Before I start a gauntlet match I always let it run for 10-15 minutes at 40/10 just to check of everything runs as it should. Then I look at the 40/10 results which gives a raw impression if the pool is balanced enough. In this case Dragon 2.5 scored 56% at 40/10 which gave me confidence knowing that a time control of 40/40 would favor the lower rated engines. Meanwhile the score for Dragon 2.5 has shank to 46%.
90% of coding is debugging, the other 10% is writing bugs.
I was going to say I think you went too far, it would be tough for Dragon to approach 50%, but as I write this it is dead even!
Before I start a gauntlet match I always let it run for 10-15 minutes at 40/10 just to check of everything runs as it should. Then I look at the 40/10 results which gives a raw impression if the pool is balanced enough. In this case Dragon 2.5 scored 56% at 40/10 which gave me confidence knowing that a time control of 40/40 would favor the lower rated engines. Meanwhile the score for Dragon 2.5 has shank to 46%.
Actually this is a good result, its performance rating vs. the 2700-2730 field was about 2795, and as of now it is running at about 2820 against the 2842 field. So overall it is performing a bit above the 2800 level giving knight odds. Probably this would be about 2700 at "true" knight odds, without the advantage of the relatively favorable first 100 positions from the set. I would estimate that the GM level CCRL engines play about as well at your bullet time control as comparably rated (comparing CCRL Blitz ratings to human FIDE ratings) humans would at an average blitz time control like 3' + 2". So perhaps that means that Dragon 2.5 could give knight odds in blitz to human GMs rated around 2700. That's probably about right; Dragon 2 was a bit plus at about that time control against GM Alex Lenderman, who was about 2632 FIDE standard and about 2650 FIDE blitz at the time.
Rebel wrote: ↑Sat Sep 25, 2021 8:40 am
One thing I still want to know, what happens if we increase the time control from 40/40 to 40/120, so CCRL 40/2.
Previous Dragon 2.5 result vs elo 2715 pool = 61.4% at 40/40
Rebel wrote: ↑Sat Sep 25, 2021 8:40 am
One thing I still want to know, what happens if we increase the time control from 40/40 to 40/120, so CCRL 40/2.
Previous Dragon 2.5 result vs elo 2715 pool = 61.4% at 40/40
KNIGHT odds match Komodo Dragon 2.5 vs a pool of 2715 elo rated engines
Time Control : Time control : 40/120
Games : 700
Results from file all.pgn:
No. Name Win Draw Loss Unf. Score Games %
--------------------------------------------------------------
1 Komodo-Dragon 2.5 +324 =106 -270 *0 377.0 700 53.9%
2 Benjamin 1.0 +48 =23 -29 *0 59.5 100 59.5%
3 ProDeo 2.2 +45 =21 -34 *0 55.5 100 55.5%
4 k2 099 +45 =11 -44 *0 50.5 100 50.5%
5 Zahak 5.0 +42 =4 -54 *0 44.0 100 44.0%
6 Dumb 1.8 +36 =15 -49 *0 43.5 100 43.5%
7 Velvet 1.2.0 +29 =19 -52 *0 38.5 100 38.5%
8 Fruit 2.1 +25 =13 -62 *0 31.5 100 31.5%
Total Games: 700
White Wins: 324 (46.3%)
Black Wins: 270 (38.6%)
Draws: 106 (15.1%)
Unfinished: 0 (0.0%)
Estimated ratings for this elo 2715 pool
# PLAYER : RATING POINTS PLAYED (%)
1 Benjamin 1.0 : 2807.0 59.5 100 60
2 ProDeo 2.2 : 2778.3 55.5 100 56
3 k2 099 : 2743.1 50.5 100 51
4 Komodo-Dragon 2.5 : 2739.6 377.0 700 54
5 Zahak 5.0 : 2697.3 44.0 100 44
6 Dumb 1.8 : 2693.8 43.5 100 44
7 Velvet 1.2.0 : 2657.5 38.5 100 39
8 Fruit 2.1 : 2603.4 31.5 100 32
40/40 : 61.4%
40/120 : 53.9%
Larry, at which time control do you usually play GM's?
When we play "par" GMs at knight odds (around the 2500 FIDE rating required for the title normally) we use the now standard FIDE Rapid time control of 15' + 10" increment. When we play top GMs we use that TC for smaller handicaps, but for knight odds it would be blitz, most likely 5' + 1". What I've discovered with Dragon giving knight odds is that Dragon doesn't benefit at all from thinking longer than what it could do at 2' + 1" blitz, whereas the opponent benefits greatly. With too much time to think Dragon just sees losses everywhere and doesn't know what to do. This seems to be true against engines and against humans. So knight odds performance will really start to drop if you go from blitz out to Rapid.
Rebel wrote: ↑Sat Sep 25, 2021 8:40 am
One thing I still want to know, what happens if we increase the time control from 40/40 to 40/120, so CCRL 40/2.
Previous Dragon 2.5 result vs elo 2715 pool = 61.4% at 40/40
KNIGHT odds match Komodo Dragon 2.5 vs a pool of 2715 elo rated engines
Time Control : Time control : 40/120
Games : 700
Results from file all.pgn:
No. Name Win Draw Loss Unf. Score Games %
--------------------------------------------------------------
1 Komodo-Dragon 2.5 +324 =106 -270 *0 377.0 700 53.9%
2 Benjamin 1.0 +48 =23 -29 *0 59.5 100 59.5%
3 ProDeo 2.2 +45 =21 -34 *0 55.5 100 55.5%
4 k2 099 +45 =11 -44 *0 50.5 100 50.5%
5 Zahak 5.0 +42 =4 -54 *0 44.0 100 44.0%
6 Dumb 1.8 +36 =15 -49 *0 43.5 100 43.5%
7 Velvet 1.2.0 +29 =19 -52 *0 38.5 100 38.5%
8 Fruit 2.1 +25 =13 -62 *0 31.5 100 31.5%
Total Games: 700
White Wins: 324 (46.3%)
Black Wins: 270 (38.6%)
Draws: 106 (15.1%)
Unfinished: 0 (0.0%)
Estimated ratings for this elo 2715 pool
# PLAYER : RATING POINTS PLAYED (%)
1 Benjamin 1.0 : 2807.0 59.5 100 60
2 ProDeo 2.2 : 2778.3 55.5 100 56
3 k2 099 : 2743.1 50.5 100 51
4 Komodo-Dragon 2.5 : 2739.6 377.0 700 54
5 Zahak 5.0 : 2697.3 44.0 100 44
6 Dumb 1.8 : 2693.8 43.5 100 44
7 Velvet 1.2.0 : 2657.5 38.5 100 39
8 Fruit 2.1 : 2603.4 31.5 100 32
40/40 : 61.4%
40/120 : 53.9%
Larry, at which time control do you usually play GM's?
When we play "par" GMs at knight odds (around the 2500 FIDE rating required for the title normally) we use the now standard FIDE Rapid time control of 15' + 10" increment. When we play top GMs we use that TC for smaller handicaps, but for knight odds it would be blitz, most likely 5' + 1". What I've discovered with Dragon giving knight odds is that Dragon doesn't benefit at all from thinking longer than what it could do at 2' + 1" blitz, whereas the opponent benefits greatly. With too much time to think Dragon just sees losses everywhere and doesn't know what to do. This seems to be true against engines and against humans. So knight odds performance will really start to drop if you go from blitz out to Rapid.
Is there an interface that I can use to play different time control for 2 engines and how exactly to do it?
I remember that in the past I used arena for unequal time control matches and also winboard but I do not find how to do it in the arena that I have now and I do not see user friendly interface for winboard for this purpose and I do not like to look again how to do it there.
