Author: Gerd Isenberg
Date: 14:43:30 03/10/03
Go up one level in this thread
On March 10, 2003 at 17:10:43, Tom King wrote: >On March 10, 2003 at 17:06:10, Gerd Isenberg wrote: > >>On March 10, 2003 at 14:49:30, Tom King wrote: >> >>>Hi all, >>> >>>One search extension which most programs seem to use is the passed pawn pushing >>>extension. When a move is made which is a passed pawn push to the 7th rank, we >>>extend the search by a ply. >>> >>>My programs has got away without this extension for years. Or perhaps "got away" >>>isn't right.. there have been losses which this extension might have prevented. >>> >>>Anyway, what do people think of this extension. Worth having? If so, what value >>>would you attach to it? >>> >>>Regards (and hi to all), >>>Tom >>>tom@silentshark.co.uk >> >>Hi Tom, >> >>i do some fractional extensions with passers. Not only if they push or capture >>foreward to the 6th or 7th rank, even if a opposite "guard" pawn in backward or >>knight distance (re)captures (but no push) from it's base rank (own 7th rank) to >>the "right" side, which establish an own passer on the 6th rank. I consider the >>number of advanced passers per side, supporting and defending pieces. A lot of >>code, specially at interiour nodes or near the root to look for further >>recuductions of determined extensions. >> >>Even if these extensions work fine in some testpositions, i requires a lot of >>effort to tune them for tournament play. Most often i do it by "gut feel". >> >>One thing i tried to control extensions with is to vary the amount of fractions >>to become one ply. I tried to vary it by search depth and by the relative amount >>of extensions so far. >> >>Regards, >>Gerd > >Hi Gerd, > >Sounds like your passed pawn extensions are very sophisticated - doesn't this >analysis slow up your search? Yes, but for nodes with depth left > 0. For leave nodes i use simpler and faster code so i believe it pays off. The final fractional depth remainder at leave nodes triggers some qsearch features, to make the qsearch width dependend on the remaining fractional depth, eg. check moves or an optional quite hint move (if any) from eval. > >I'm sure you are dead right about the test positions. I gave up with testsuites >a while back, and now just use hundreds of test games to determine whether a >change is beneficial or not. > >Regards, >Tom That's the right way to do it - but there are so many screws and switches. I still use some testpositions and during tournament play i'll hope let the heuristics fit ;-) This position was sensible to passed pawn extensions for me. After IsiChess lost the game againts Shredder in Maastricht after c3? i did some extension (passed pawn reductions but some other additional extensions) and even eval tuning to solve this - but don't ask me whether IsiChess is stronger now. [D] 8/k1p1r3/p2b3p/5ppq/1np5/6P1/1R1N1P1P/R4QK1 b - - ; am c3 Regards, Gerd
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.