Author: Andrew Williams
Date: 03:29:14 05/25/00
Go up one level in this thread
On May 25, 2000 at 03:50:38, Roberto Waldteufel wrote: > >[snip] > >>My passed pawn bonus can go _way_ over a pawn. Two connected passed pawns >>(in the middlegame) can go way over 3 pawns of positional advantage. You can >>work around the lazy eval problem, as I do the same thing. > >This is interesting - I wonder how you work around it. > My program keeps track of the highest positional score encountered so far in the current search, and uses this as the trigger margin for a lazy exit. I have a minimum for this, which is about 2 pawns. I think I ripped this idea off from crafty. >Does this mean that you test for passed pawns at the same time as computing the >material balance, but before doing other positional evaluation? This had >occurred to me as a possible solution, but I have refrained from doing it so far >for fear of the speed cost. Having witnessed some horrible losses I may have to >accept the cost and do it anyway - at least that way I can bump up the bonus to >any size I like without worrying about the lazy evaluation tolerance limits. >Maybe a good compromise would be to test only pawns on 4th-7th or 5th-7th ranks >at first, and do the rest of the pawns only if the lazy evaluation tolerance >limits are exceded? That way the test could use only half of the pawn bitboards >for each side's passed pawns, so the calculation would be a bit faster on my >32-bit hardware. > I used to use pawn-hash info to tell me how many pawns are passed and either not allow a lazy exit, or bump up the margin by an arbitrary amount per passed pawn. These things are all worth experimenting with. I also used to extend a quarter ply on a pawn push to the 6th, so long as it wasn't blockaded and there was less than a certain amount of enemy material and the promotion square wasn't covered etc etc. I took this extension out in the end. Andrew
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.