Author: Robert Hyatt
Date: 09:15:10 04/17/04
Go up one level in this thread
On April 17, 2004 at 01:44:23, Dann Corbit wrote: >On April 16, 2004 at 22:37:36, Robert Hyatt wrote: > >>On April 16, 2004 at 16:40:49, Dann Corbit wrote: >> >>>Why not probe memory-resident bitbase EGTB files for every non-pv node and >>>Nalimov (or other advanced EGTB) probes for every pv node? >> >>How do you know which is which, before the fact??? > >The node I search first will be the pv node 90+% of the time. If you go to depth=15,there are exactly _fifteen_ of those nodes. What about the other one billion??? > >>> >>>If it is not a PV node, then we don't really care what the move to make is. And >>>if we get a fail high on a bitbase probe, it will get researched as a PV node >>>anyway. >>> >>>It seems like it would cut EGTB probes to 1/30 or so of probing all the time, >>>and the probes would always be where they are needed the most. >> >>Searching the PV takes over 1/2 of the time for Crafty, in most cases. Your >>savings would be far lower than 1/30. Probably not even 1/2.. > >I had not thought about that. It seems on the surface that there should be a >large savings since most nodes are not pv nodes. > >>>Have others already tried this?
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.