As playing engines for years I agree that we (humans) have an impression about a playing style. Unfortunately I usually get disapointed sooner or later because trying to "understand" top engines "playing style" is in vain ...oreopoulos wrote:As a player that just uses engine for analysis and not like horses that compete my observations are...
Thanks, it seems Houdini 4 is beating consistently everything at this TC. It is tested at Contempt=1, right?Houdini wrote:People at the Rybka Forum asked for the individual match results, I might as well also publish them here.Houdini wrote:Ingo Bauer was so kind to run the IPON rating list matches for the new Houdini 4.
The result is a +39 Elo increase compared to Houdini 3.
The new complete IPON rating list is given below.
Note that in the following list the Elo scale is calibrated differently, setting Houdini 3 at 3000 points.
Code: Select all
Houdini 4 - Komodo 6 (2971) 89.5 - 60.5 59.67% Perf=3039 Houdini 4 - Stockfish 4 (2947) 88.5 - 61.5 59.00% Perf=3010 Houdini 4 - Gull 2.2 (2908) 106.5 - 43.5 71.00% Perf=3063 Houdini 4 - Critter 1.4a (2907) 106.5 - 43.5 71.00% Perf=3062 Houdini 4 - Deep Rybka 4.1 (2882) 110.5 - 39.5 73.67% Perf=3060 Houdini 4 - Hannibal 1.4a (2798) 121.0 - 29.0 80.67% Perf=3046 Houdini 4 - Chiron 1.5 (2779) 125.5 - 24.5 83.67% Perf=3062 Houdini 4 - Protector 1.5.0 (2771) 129.0 - 21.0 86.00% Perf=3086 Houdini 4 - Naum 4.2 (2768) 131.5 - 18.5 87.67% Perf=3108 Houdini 4 - HIARCS 14 WCSC 32b (2747) 126.0 - 24.0 84.00% Perf=3035 Houdini 4 - Deep Shredder 12 (2730) 135.5 - 14.5 90.33% Perf=3118 Houdini 4 - Jonny 6.00 (2729) 130.0 - 20.0 86.67% Perf=3054 Houdini 4 - Deep Sjeng c't 2010 32b (2713) 126.0 - 24.0 84.00% Perf=3001 Houdini 4 - Spike 1.4 32b (2707) 131.5 - 18.5 87.67% Perf=3047 Houdini 4 - spark-1.0 (2695) 138.5 - 11.5 92.33% Perf=3127 Houdini 4 - Deep Junior 13.3 (2677) 137.0 - 13.0 91.33% Perf=3086 Houdini 4 - Booot 5.2.0 (2674) 138.5 - 11.5 92.33% Perf=3106 Houdini 4 - Quazar 0.4 (2665) 139.5 - 10.5 93.00% Perf=3114 Houdini 4 - Zappa Mexico II (2652) 139.0 - 11.0 92.67% Perf=3092 Houdini 4 - Toga II 3.0 32b (2643) 138.5 - 11.5 92.33% Perf=3075 2488.5 - 511.5 82.95% Perf=3042
This is all default except IPON specific settings.Laskos wrote:
Thanks, it seems Houdini 4 is beating consistently everything at this TC. It is tested at Contempt=1, right?
Considering our 5'+3" results I doubt that. Actually Houdini 4 is rated 3117 while performance of StockyDD is ~3060-3070. So around +40-50 points in f/o Houdini, which is definetely not "to close to call" especially on this high level...lkaufman wrote:.... while at standard chess they are too close to call.
TCEC is a different beast than any of the rating lists. Lots of cores, lots of time to think for the engines.IGarcia wrote: Martin nTCEC2 was a great competition, but without offense, is not statistically representative.
There is no standard chess. TC without hardware is meaningless.lkaufman wrote:Based on the results posted here so far, the fairest thing to say about Houdini 4 and StockfishDD (or any November version) is that Houdini is clearly stronger in blitz, while at standard chess they are too close to call. A. Huerga's new LTC list, based on at least 600 games per engine, has Stockfish 5 elo points ahead of Houdini 4, well within the error margin. These results do show that Stockfish scales better than Houdini, so I would probably use SF over Houdini for analysis (i.e. for second opinion after Komodo!), but the jury is still out.
Hello Ingo, is it planned to test Stockfish DD in 5m+3s ?IWB wrote:As playing engines for years I agree that we (humans) have an impression about a playing style. Unfortunately I usually get disapointed sooner or later because trying to "understand" top engines "playing style" is in vain ...oreopoulos wrote:As a player that just uses engine for analysis and not like horses that compete my observations are...
Bye
Ingo
By standard chess I mean time limits that FIDE would rate, such as 90' + 30" or longer. It is very clear that Houdini's superiority over Komodo and Stockfish at blitz does not translate to superiority at these much longer time limits. If you believed that time limit doesn't matter (other than the spread of the ratings), why would CEGT and CCRL bother to test at short and long time controls?Wolfgang wrote:Considering our 5'+3" results I doubt that. Actually Houdini 4 is rated 3117 while performance of StockyDD is ~3060-3070. So around +40-50 points in f/o Houdini, which is definetely not "to close to call" especially on this high level...lkaufman wrote:.... while at standard chess they are too close to call.
I admit that with longer TC (40/20, 40/40 or even 40/120) the differences will decrease.
It WAS planned after H4!Vinvin wrote: Hello Ingo, is it planned to test Stockfish DD in 5m+3s ?
Sure, its more like you say and great fun and interesting.Martin Thoresen wrote: TCEC is a different beast than any of the rating lists. Lots of cores, lots of time to think for the engines.
Closer to the user behavior when they use an engine for analysis? Yes.
Statistically comparable to blitz rating lists? No.
It was planned and testing now, or... "WAS" means now is canceled?IWB wrote:It WAS planned after H4!Vinvin wrote: Hello Ingo, is it planned to test Stockfish DD in 5m+3s ?
Inog ?Bye
Inog
Funny I many times type my name Igancio, nasty bug, seems im not the only one.