Author: scott farrell
Date: 07:54:28 11/04/03
Go up one level in this thread
On November 04, 2003 at 08:33:49, Tord Romstad wrote: >On November 04, 2003 at 06:42:56, scott farrell wrote: > >>Consider the position below, just moving into the qSearch my engine tries the 2 >>checks, and RxP. >> >>The checks lead to a qsearch for white that looses the rook. > >Does this mean that you search checks like Rb1+ and Ra2+ in your qsearch? >I did this before, but removing checks which lose material from the qsearch >was a big improvement for me. > >>The RxP leads to a qsearch for white of the promotion. >> >>Clearly it doesnt like any of these moves, so it just takes the stand-pat at the >>node it is at. As such it is horizoning the fatalities in the qsearch. >> >>Is this correct? > >Yes, I think so. > >>How many people use stand-pat ala crafty? Who uses what else? > >My engine does like Crafty. > >>even though my engine has the pawn promotion in the qSearch, the stand-pat >stops it from playing it. > >I try to solve such problems with my evaluation function. A passed pawn >on the 7th which can safely advance and promote (I check this by calling >my SEE from within the evaluation function) is given a huge bonus. My >engine will not want to stand pat in the position you use as an example, >because the static eval returns a very bad score (-5.43). >> >>A 1 depth search from this node, gives me 75 nodes, and the score drops by 5 for >>black. >> >>One solution I can see is to staticly evaluate the rook as lost, so it wont >>stand-pat at the node. My evaluate function doesnt really consider who's move it >>is, to see the rook is undefended and under attack, and sideToMove is going to >>allow its capture. I have some eval code for pinned pieces and enpris and the >>like, but low values - it would seem to me here I could statically say the Rook >>is lost. > >I don't think this is a good idea. That black's rook is hanging is not >really very relevant here. You could move the black rook to almost any >other square on the board, and black would still be lost. > >>The only other solution I can think is to let the other side make its qsearch >>moves anyway, sort of like null-move, and given they all fail-high for white, >>dont allow the stand pat or something (sort of like dont allow stand-pat during >>checks). > >I haven't tried this, but it sounds rather expensive. The best idea might >be to take a close look at your passed pawn evaluation. Two connected >passed pawns on the 7th rank should be evaluated as being worth more than >a rook, IMHO. > I agree, I already do that. I was just interested in that the stand-pat and qsearch cant find it, that you need to use evaluation. Clearly the above is anything but quiet, but your supposedly _quiet_ eval function has to guess at the dynamic. >Tord
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.