Author: Robert Hyatt
Date: 20:21:51 10/22/98
Go up one level in this thread
On October 22, 1998 at 20:12:05, Larry Coon wrote: >On October 22, 1998 at 18:50:37, Robert Hyatt wrote: > >>your only mistake is "no safety outlet". IE if you only generate >>"interesting moves", then you need a safety valve for when those >>"interesting mvoes" are no good. The best way is what we all do in >>the quiescence search, namely to first call evaluate(), so that you have >>a choice of either "standing pat" and accepting the current score, or else >>trying and accepting the score from one of the interesting moves, if the >>scores are better than the score you get by "doing nothing"... > >I ended up trying something like that after I posted my question. >My move generator now returns a flag that's true if there are >addditional moves that weren't returned. At the end of the >search routine, before returning alpha, I now check the flag. If >it's true and the static eval for the current position > alpha, >then I set alpha to the static eval for the current position. >(I was already calling evaluate at the start of the routine to >detect terminal positions and for a basis for finding killer >moves.) > it is more efficient to call evaluate *first*.. as this gives you a better "lower bound" to search against, rather than doing the captures and *then* choosing to stand pat. You choose to stand pat and let the captures improve the score if they can... much easier, cleaner, and will search fewer nodes... >This seems to provide a "safety outlet," as you said, but I'm >uncomfortable with how safe the safety outlet is. If all the >non-interesting moves I didn't evaluate also lead to worse >positions, then returning the static eval for the current >position will return a misleading value. > >On the other hand, I guess it's no less reliable than any other >static evaluation on a non-terminal position.... > >Was the way I did it the right way? > >Larry Coon
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.