Page 1 of 1

Fire 1.31 match.

Posted: Thu Jun 10, 2010 5:02 am
by Mark Young
Nice to be here on Open chess where people can talk openly about chess engines.

I have been testing Fire 1.31 at 10m/40moves. Testing Fire 1.31 with Ponder move = Never, Bishop pair= 70. I have come to suspect that their is a bug in the parameters in Fire 1.31. It seems that the Parameters in Fire 1.31 for contempt and Bishop pair have some kind of flaw. I have come to suspect that the setting of 50 for both Bishop Pair and contempt are really the 0 point. This is why the default contempt for Fire 1.31 is set at 64. This would in true terms equal (64-50=14) 14 as the true contempt settings. And the default Bishop Pair at 50 is really 0. This is why Fire 1.31 will not favor the bishop pair, and why when you lower the contempt factor below 50 you get really strange eval reading, or the engine just crashes if you set cotempt to low. This is only a working theory that I am testing.

WIth the new Fire 1.31 settings of Bishop pair = 70, Fire 1.31 is now #1 on my ratings list at 1/1inc, 3/1inc and 5/1inc with default setting Fire 1.31 was #5 on my ratings list, I am testing to see if the new setting still hold ture at slower time controls of 10m/40moves.

Gauntlet match to 100 games. (200 game in total)

Fire 1.31 w32 - Deep Rybka 4 w32 16.5 - 13.5 +7/=19/-4 55.00%
Fire 1.31 w32 - Houdini 1.01 w32 4_CPU 13.5 - 14.5 +2/=23/-3 48.21%

Re: Fire 1.31 match.

Posted: Thu Jun 10, 2010 10:20 am
by Volen
AMP Phenom 9850 2.5x4, W7x64, pondef off, F12 GUI, Perfect 2009, 512mb hash, 3-4-5 TB, 100 5'+0 games.
Image
Image

Seems like non-popcnt version work better on my PC

Re: Fire 1.31 match.

Posted: Thu Jun 10, 2010 10:17 pm
by Jarkko
Are you sure? The error bars seem to be overlapping.