Gothmog with Sel 5 stronger?

[AndyC]

Geschrieben von:/Posted by: AndyC at 02 April 2004 11:53:37:

AMD Barton 3200+. Engines using 256 MBs ram for hash (and 16MBs ram for pawn/EGTB hash if needed) where possible or max if less than 256MB. Own opening books except Gothmog using an Arenabook I created. 20 mins + 5 sec under Arena GUI.


Gothmog 0.4.7 - 5   - Abrok 5.0                  1.5 - 2.5
Gothmog 0.4.7 - 5   - Amyan 1.592                3.5 - 0.5
Gothmog 0.4.7 - 5   - AnMon 5.30                 1.5 - 2.5
Gothmog 0.4.7 - 5   - Comet B.67                 3.0 - 1.0
Gothmog 0.4.7 - 5   - Delfi 4.4                  3.0 - 1.0
Gothmog 0.4.7 - 5   - El Chinito 3.25            2.0 - 2.0
Gothmog 0.4.7 - 5   - Francesca MAD 0.0.9        3.0 - 1.0
Gothmog 0.4.7 - 5   - Gandalf 432h               1.0 - 3.0
Gothmog 0.4.7 - 5   - Green Light Chess 3.00     1.5 - 2.5
Gothmog 0.4.7 - 5   - King of Kings 2.52         2.0 - 2.0
Gothmog 0.4.7 - 5   - Leila 0.53h                3.0 - 1.0
Gothmog 0.4.7 - 5   - PatzerUCI Leiden           3.0 - 1.0
Gothmog 0.4.7 - 5   - Quark 2.05                 2.5 - 1.5
Gothmog 0.4.7 - 5   - SOS 4 for Arena            1.5 - 2.5
Gothmog 0.4.7 - 5   - SlowChess 2.89             3.0 - 1.0
Gothmog 0.4.7 - 5   - SmarThink 0.17a            1.5 - 2.5
Gothmog 0.4.7 - 5   - WildCat 2.79               4.0 - 0.0
Gothmog 0.4.7 - 5   - Yace Paderborn             2.0 - 2.0


Again I preface discussion with the sample size being small. With a selectivity setting of 5 it scores a total of 42.5/72. Compare this to earlier.
Total Scores
Selectivity 4 - 34.5
Selectivity 6 - 38.5
Selectivity 7 - 35.0
If the results are accurate then Gothmog with a selectivity of 5 would be some 30 ELO points stronger again c.f. Gothmog with a selectivity of 6. However I'm not sure that tweaking just one of a program's parameters could improve it that much (60-65 ELO points compared to it's original selectivity setting)? Perhaps someone with longer experience in testing could comment on whether such gains are possible?
Also Tord tells me that in 100 blitz games vs Crafty the two settings (5 & 6) score almost identically. All this points to a need for further testing I think.
Am going to run the more games comparing selectivity of 5 and 6 but am going to be away for awhile. I'd urge some other testers to try the selectivity = 5 compared with 6 and report the results .

[Tord Romstad]

Geschrieben von:/Posted by: Tord Romstad at 02 April 2004 12:31:29:
Als Antwort auf:/In reply to: Gothmog with Sel 5 stronger? geschrieben von:/posted by: AndyC at 02 April 2004 11:53:37:

Also Tord tells me that in 100 blitz games vs Crafty the two settings (5 & 6) score almost identically.
All this points to a need for further testing I think.
Am going to run the more games comparing selectivity of 5 and 6 but am going to be away for awhile. I'd urge some other testers to try the selectivity = 5 compared with 6 and report the results .

Yes. The exact results were:

Gothmog 0.4.7 (sel 6) - Crafty 19.9: 50.5-49.5
Gothmog 0.4.7 (sel 5) - Crafty 19.9: 49.0-51.0

No significant difference, in other words.
Before somebody starts claiming that Gothmog seems to be as strong as Crafty,
I'd like to point out that the games were played without pondering, and that
Crafty's time management is designed to work with pondering enabled. Bob
has always made clear that Crafty is much weaker without pondering, and that
results in such games should not be used to estimate Crafty's strength.
Definitely. I will of course do some further tests myself, too.
If somebody volunteers to do such tests, I would of course be very grateful.
But in serious tournaments, I still recommend that selectivity=6 is used.
Tord