Rhetoric 1.4 64-bit Gauntlet for CCRL 40/40

[Graham Banks]

PGN - http://kirill-kryukov.com/chess/discuss ... p?id=32533

[asanjuan]

Thanks Graham.
It is very surprising to see that in the ccrl 40/40 list the improvement is about +10 elo, but the CEGT lists showed a +60 elo improvement in 40/20, and +37 in 40/4.
It is very confusing...

Has anyone tested Rhetoric 1.4 for the CCRL 40/4? I'm missing Rhetoric there. I hope to see something like a coherent result somewhere.

Regards.

[Graham Banks]

Thanks Graham.
It is very surprising to see that in the ccrl 40/40 list the improvement is about +10 elo, but the CEGT lists showed a +60 elo improvement in 40/20, and +37 in 40/4.
It is very confusing...

Has anyone tested Rhetoric 1.4 for the CCRL 40/4? I'm missing Rhetoric there. I hope to see something like a coherent result somewhere.

Regards.

The test results are what they are. I'm sorry that they didn't meet your expectations.

I'm sure that if Rhetoric 1.4 isn't in our 40/4 list already, it won't be missing for long.

Good luck in my Division 3 tournament starting later this week.

Graham.

[asanjuan]

Thanks Graham.
It is very surprising to see that in the ccrl 40/40 list the improvement is about +10 elo, but the CEGT lists showed a +60 elo improvement in 40/20, and +37 in 40/4.
It is very confusing...

Has anyone tested Rhetoric 1.4 for the CCRL 40/4? I'm missing Rhetoric there. I hope to see something like a coherent result somewhere.

Regards.

The test results are what they are. I'm sorry that they didn't meet your expectations.

I'm sure that if Rhetoric 1.4 isn't in our 40/4 list already, it won't be missing for long.

Good luck in my Division 3 tournament starting later this week.

Graham.

Completely agree, of course.
The point is not that it didn't meet my expectations, it is that I see three completely different results. In previous versions, the gain in the strenght in almost every list were in line with my own tests, but this time they are all completely different!
I can't see any correlation.

It makes me think in a bug... or it is that more games are needed.
Time will tell.
Thank you very much Graham.

[pedrox]

Hi Alberto,

Here the Hispanic list, the games here are 2 +2 on a Buldozer computer and each engine playing with her own book (Rhetoric with performance.bin). I have the impression that maybe the gain is greater with the depth reached by the engine, but I have no data.

Code: Select all

Rank Name                       Elo    +    - games score oppo. draws 
   1 Gaviota 1.0 x64           2916   33   31   464   83%  2646   21% 
   2 Gaviota 0.86 x64          2843   39   37   334   82%  2571   21% 
   3 Dirty 20APR2013 pop64     2784   29   28   464   69%  2642   27% 
   4 Dirty 20APR2013 pop64 ne  2772   37   36   280   68%  2634   28% 
   5 Chronos 1.99b x64         2764   23   22   918   75%  2487   23% 
   6 Gaviota 0.85.1 x64        2762   47   44   341   87%  2253   14% 
   7 Dirty 5Sep2012 x64        2752   33   32   536   82%  2375   18% 
   8 Dirty 20APR2013 x64       2750   35   34   310   68%  2615   25% 
   9 Atlas 3.60 x64            2736   27   27   534   66%  2591   25% 
  10 Andscacs 0.64 pop64       2729   32   31   414   69%  2564   27% 
  11 Godel 3.4.9 x64           2712   29   29   464   63%  2596   25% 
  12 Dirty 23Mar2012 x64       2700   55   51   288   87%  2179   10% 
  13 Rhetoric 1.4 x64          2698   29   29   404   56%  2651   30% 
  14 Rhetoric 1.2 x64          2677   34   34   350   62%  2568   25%
  15 Alfil 13.1 x64            2654   22   22   811   58%  2582   22% 
  16 Atlas 3.50 x64            2654   43   42   230   70%  2484   24% 
  17 Rodin 7.0                 2648   31   31   414   59%  2568   22% 
  18 Alfil 12                  2633   44   43   323   77%  2253   14% 
  19 Dirty 23Mar2012 w32       2630   51   49   262   82%  2204   11% 
  20 Godel 2.3.7 x64           2626   26   26   592   63%  2512   23% 
  21 Atlas 3.25 x64            2614   29   29   553   66%  2409   16% 
  22 Rodin 6.0                 2611   31   31   398   54%  2571   23% 
  23 DanaSah 5.07              2606   26   27   543   50%  2602   22% 
  24 Rodin 5.1                 2590   30   30   438   59%  2515   22% 
  25 Amyan 1.72                2588   21   21   925   55%  2491   21% 
  26 DanaSah 5.00              2585   29   29   463   61%  2493   26% 
  27 Andscacs 0.62u pop64      2576   36   36   310   51%  2563   23% 
  28 DanaSah 5.06              2568   41   41   229   60%  2484   25% 
  29 Andscacs 0.60l pop64      2546   38   38   259   55%  2498   24% 
  30 DanaSah 4.88              2541   44   43   311   72%  2226   14% 
  31 Pepito 1.59.2 x64         2537   21   21   948   50%  2487   20% 
  32 Cerebro 3.03d             2522   22   22   936   49%  2488   18% 
  33 Rodin 4.0                 2515   45   44   302   72%  2211   13% 
  34 Bruja 1.9.1 x64           2501   20   21  1076   50%  2465   18% 
  35 Rhetoric 1.1 x64          2499   38   38   260   49%  2501   26%
  ... 

[asanjuan]

Hi Alberto,

Here the Hispanic list, the games here are 2 +2 on a Buldozer computer and each engine playing with her own book (Rhetoric with performance.bin). I have the impression that maybe the gain is greater with the depth reached by the engine, but I have no data.

Hi Pedro,
some people reported me a similar behaviour. At longer time controls, seem to perform better. It could be an effect derived from the learning process, or a small change in the search. Now Rhetoric uses the board's attack information to not to prune moves from threatened pieces (LMP).

It is a costly operation, and I have always thought that it should be better at low TC, but experience is showing just the opposite.

Anyway, your list shows a +22 elo improvement, so what I have in general:
- CEGT 40/20: +60 (800 games)
- CEGT 40/4: +37 (1000 games)
- CCRL 40/40: +8 ? (246 games)
- Hispanic: (similar to ~40/4?): +22 (404 games)

Still, don't know how stronger is the new version. I have to be patient and wait.

[asanjuan]

Thanks Graham.
It is very surprising to see that in the ccrl 40/40 list the improvement is about +10 elo, but the CEGT lists showed a +60 elo improvement in 40/20, and +37 in 40/4.
It is very confusing...

Has anyone tested Rhetoric 1.4 for the CCRL 40/4? I'm missing Rhetoric there. I hope to see something like a coherent result somewhere.

Regards.

I am certainly going to test it at some point but at the moment I can't find the CPU resources to do it. The only thing I can say for sure is that it has already found its way to my testing queue.

Definitely, I have to give credit to the CCRL testing group in my web page!