Author: Robert Hyatt
Date: 13:37:53 07/09/02
Go up one level in this thread
On July 09, 2002 at 16:14:31, Uri Blass wrote: >On July 09, 2002 at 12:48:18, Robert Hyatt wrote: > >>On July 09, 2002 at 03:40:31, Christophe Theron wrote: >>Here is some sample data from crafty: >> >>nodes searched: 9.25M >>nodes searched below a NULL move: 7.75M >>null_move searches failing high: 1.02M >>null_move searches failing low: .235M >> >>Another position: >> >>nodes searched: 15M >>nodes searched below a NULL move: 5.5M >>null_move searches failing high: 2.0M >>null_move searches failing low: .5M >> >>That is why I said "this is not about a few percentage points." >> >>First position researched with R=4, just for fun: >> >>Nodes: 4.7M >>below NULL: 3.3M >>fail high: .624M >>fail low: .138M >> >>Going from R=2~3 to R=4 reduced the search time by 50%. > >It is interesting to see direct comparison between R=3 and R=4. > >I did direct comparison in few test positions between recursive R=2 and >recursive R=3 and >there are positions when the time is reduced by more than 50%. > >I now run the GCP test suite(5 minutes per move) for recursive R=3. > >I am going to compare it later also with R=3 not recursive and with other >options. > >Uri beware of running tactical tests only. They tend to highlight the utility of null-move since so many branches are horribly losing. I use a combination of positions with some tactics, some quiet, some opening, some endgames...
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.