Author: John Scalo
Date: 16:29:59 11/20/97
Go up one level in this thread
On November 20, 1997 at 19:12:54, Edward Screven wrote: >On November 20, 1997 at 18:49:17, John Scalo wrote: > >>Q- Is it acceptable to run through the qsearch without ever evaluating >>the position? In other words only the material balance is considered. It >>sure is a lot faster and most positional eval's don't include a term >>worth more than half a pawn. But there are some obvious dangers such as >>ending up with a king in the middle of the board open to attack, etc. I >>suppose it's a tradeoff of speed/quality of play. Is the speed worth it? > >i tried this once. my program not only played worse, but it scored >lower on test suites. but you should try it for yourself -- it should >be easy to implement. I'm actually moving in the opposite direction -- I currently do not consider the positional eval and I'm considering doing so, but not looking forward to the speed hit. > >i think a better direction is to work on forward pruning to reduce your >branching factor. for example, do you use a static exchange evaluator >to nick losing captures? how about razoring and futility cutoffs? >also, you might try lazy evaluation. > Yes in some part to all of these, except for the s.e.e. I currently sort by capture net gain (i.e. pawn takes king goes first, king takes pawn goes last). I'm looking for pointers on doing a s.e.e. for better sorting but my use of bitboards is currently limited and I understand this is a critical tool in making a fast s.e.e. Thanks, -j
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.