Author: Joachim Rang
Date: 03:53:00 10/14/05
Go up one level in this thread
On October 14, 2005 at 05:54:56, Uri Blass wrote: >On October 14, 2005 at 05:42:07, Joachim Rang wrote: > >>On October 14, 2005 at 02:44:30, Gian-Carlo Pascutto wrote: >> >>>On October 13, 2005 at 23:26:01, K. Burcham wrote: >>> >>>> >>>>It takes Fruit 2.2 7 minutes to play e5 >>>> >>>> [D] rn1q1k1r/1p2nppp/p2p1b2/2pP4/4PB2/2N2B2/PPP3PP/R2Q1RK1 w - - 0 1 >>>> >>>>Fruit 2.2: >>>>Athlon 64 >>>>2200 mhz >>>>432 hash >>> >>>Fruit may be "sceptical" about white's attacking chances on blacks badly stuck >>>king. >> >> >>Is that a sting on our marketing bubble? ;-) >> >>The position is already quite good for white here and as the comparison to Fruit >>2.1 shows the more aggressive pruning of Fruit 2.2 seems to hurt in this >>position. >> >>Joachim > >I wonder if the aggresive pruning of Fruit2.2(history threshold=70 in Fruit2.2 >relative to history threshold=60 in fruit2.1) is really productive at long time >control. > >Is this change in the parameters based only on blitz games or also on long time >control games? > >Tha fact that fruit2.2 can avoid some mistakes faster with history threshold=50 >is not a proof that smaller threshold is better but at least it suggest testing >smaller history threshold at long time control. > >As I showed in the past Fruit2.2 could avoid some mistakes including a losing >mistake against Deep Sjeng in WBEC with smaller history threshold. > >Uri You are right that the tuning is based on blitz-performance. However unless we can prove it statistically that smaller threshold is better in longer games we won't change it. That needs 400+ games on long time control and is just too much work. Btw we are refining history pruning anyway so tuning the threshold now before we make refinemenst is not useful. Joachim
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.