Author: Andrew Dados
Date: 18:38:49 12/13/00
Go up one level in this thread
On December 13, 2000 at 20:45:14, Scott Gasch wrote: >Hi, > >I had an idea recently about improving the ordering of non-captures. Up until >now I have been using a history heuristic (with 2 killers). > >The idea is to try non-captures that threaton the last piece moved, flee from >the last moved piece, or whose "footprint" overlaps the footprint of the last >move before their peers. By footprint I mean the set of squares a piece of type >T in square X can move to if nothing else is on the board. > >Doing some tests on different positions I decided to throw away the "flees" >case... I am skeptical of its soundness and it always grew the tree... >sometimes quite severly. > >The other two (overlapping footprints and attacks the last moved piece) both >grow the tree anywhere from slightly to greatly. I tried different bonus >weights, bonus + history and bonus alone, etc... and the best I ever got was >trees of equal size in _some_ positions. > >Despite this fact I still think there might be some merit in this idea. Surely >its better for move ordering to try, for example, threatoning pawn pushes before >quiet ones or moving a to support a square attacked by the opponent's last move >before moving randomly... My question to you guys is if anyone else is doing >anything similar and what have your results been like? > >Thanks, >Scott If you nullmove, try some simple trick, which works great for me in middlegame. When your nullmove fails low, check for refutating move (best move on next ply). If it is a capture, then try all 'escape moves' for threatened piece even before history heuristics. And maybe don't update killers/history when such an 'escape' move produces cutoff. Or maybe make a separate history for 'escape' moves... -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.