Author: Robert Hyatt
Date: 08:51:33 06/26/04
Go up one level in this thread
On June 26, 2004 at 00:15:18, Dann Corbit wrote: >On June 25, 2004 at 23:50:44, Robert Hyatt wrote: > >>On June 25, 2004 at 22:19:41, Dann Corbit wrote: >> >>>On June 25, 2004 at 21:24:59, Robert Hyatt wrote: >>> >>>>On June 25, 2004 at 16:36:19, Stuart Cracraft wrote: >>>> >>>>>Just a note that I got my hashing working in a >>>>>quick hack program I've put together for other >>>>>reasons and it was nasty even though I've been >>>>>through it over and over through the decades. >>>>>I empathize with other hashers who have felt or >>>>>are feeling hashed. >>>>> >>>>>It is well worth it. A typical 8 ply search early >>>>>in the game might reduce 50%+ in total nodes searched >>>>>and 50%+ in total time. >>>>> >>>>>These are dwarfed by null move's effect though in >>>>>the same positions (90% and 90%). >>>>> >>>>>So my question is, these are well-known methods to >>>>>substantially reduce the number of nodes and amount >>>>>of time for most searches -- I wonder if there is >>>>>anything else that is as large and as comparable >>>>>at these large 50%/90% types of reductions. >>>>> >>>>>Stuart >>>> >>>> >>>>another 50% or more. Limit your q-search captures so you don't look at silly >>>>captures that obviously just lose material. >>> >>>How is that different from an alpha/beta search of captures only? >> >> >>Don't look at _all_ captures. Only captures that appear to have a chance of >>producing something other than a fail-low (alpha cutoff)... > >Do you determine the probability of a good capture with SEE? >Otherwise it seems you might miss obvious stuff like this: >[D]1r2k1r1/pbppnp1p/1b3P2/8/Q7/B1PB1q2/P4PPP/3R2K1 w - - bm Qxd7+; Yes, but _only_ in the q-search. The q-search has so many errors in it where the best move is _not_ a capture, that the SEE errors introduced don't influence the final score at all...
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.