Fire 1.31 on top.

[mwyoung]

I have been a bit disappointed with Fire 1.31. It is a super strong program, but it would always get ground down by Houdini or earlier version of Firebird, or other strong programs from time to time. After looking at Fire 1.31 losses, it became clear to me that Fire 1.31 under valued the bishops. I tried tuning the bishop and knight settings, but I could not seemed to fix the problem.

Then I started to attack the problem more systematically and change and test only one parameter at a time. After a little trial and error, one simple change seemed to have fixed Fire 1.31 problem of the under value Bishops.

The change is simple, but it took some time to find and test. After this change Fire 1.31 has moved from number 5 on my rating list to number 1 and now tops Houdini.

I am still testing the setting, but the results have been so good I thought I would share the setting for others to try.

Settings all Default
Except BP=70, Ponder hit never.

I tested 50, 55, 60, 65 with little change in playing strength. When I tested BP=70 Fire 1.31 tested much better and came to the top of my ratings list at Fast time controls 1, 3, 5 min games +1 sec incs.

I am not sure BP=70 is the best setting.75 or 80 could ever be better, but I have not tested those settings.

I am currently testing at longer time controls against the 2 strongest program other then Fire 1.31. Houdini 1.01 and Deep Rybka 4 TC 3 100 150. To see if Fire's new settings work at longer time controls, I have only been testing at longer time controls for a day, so I don't have a bunch of games, but the new settings seem to be working.

Gauntlet match to 100 games, TC: 10m/40 , Book: GM 2500+ limited to 12 moves. HT=256, TB=5 man for Rybka, and Fire. Q6600 quad core.

Current match results at 10m/40:


Fire 1.31 w32 - Deep Rybka 4 w32......................9.0 - 5.0....... +5/=8/-1 64.29%

Fire 1.31 w32- Houdini 1.01 w32 4_CPU.............6.0 - 6.0..... +1/=10/-1 50.00%

(games 200).

[govert]

BP

I presume you mean "Bishop Pair"?

[mwyoung]

Yes .

Bishop Pair = 70

Move on ponderhit = Never.


Update on testing....

Fire 1.31 w32- Deep Rybka 4 w32.........................9.5 - 5.5 +5/=9/-1 63.33%
Fire 1.31 w32 - Houdini 1.01 w32 4_CPU..............7.5 - 6.5 +2/=11/-1 53.57%

[M ANSARI]

That is way way way too few games to make a judgement either way.

[govert]

That is way way way too few games to make a judgement either way.

I think Mark already made that pretty clear.

[mwyoung]

If 12 game is enough to decide the Human World Chess Champion. It is enough to decide the best computer settings.

I guess I was not clear.....

[notyetagm]

Yes .

Bishop Pair = 70

Move on ponderhit = Never.


Update on testing....

Fire 1.31 w32- Deep Rybka 4 w32.........................9.5 - 5.5 +5/=9/-1 63.33%
Fire 1.31 w32 - Houdini 1.01 w32 4_CPU..............7.5 - 6.5 +2/=11/-1 53.57%

Very nice work.

Looking forward to what the final conclusion is after more testing.

[mwyoung]

I think I have found the cause of Fire 1.31 strange behavior in the engine settings. Will explain this in new post.