Author: Robert Hyatt
Date: 05:39:17 03/09/00
Go up one level in this thread
On March 08, 2000 at 22:57:08, Sanjiv Karnataki wrote: >Hi, > >I noticed that in crafty all moves except the first move in the root move list >are sorted by the node count so that after the best move from the previous >iteration all other moves are in a descending order of node count searched in >the previous iteration? > >Am i correct in reading the code? and what is the advantage of doing this? > The idea is this: If you watch the search, on occasion you will notice that after searching the first move, the rest whiz by quickly, except maybe for one. This one takes a bit longer. The next iteration it takes even longer to dismiss it, and finally it actually becomes a new best move and replaces the best move at the root. The additional time for that one move is a good hint that it might eventually become the best move. By sorting at the root using the node count, these 'potential new best moves' filter to the top of the move list so that they are seen quicker... It is something we tried in Cray Blitz a long time ago, and it seemed to work well. I have used it ever since. >Currently in my program, I keep the root move sorted based on the score returned >for each move from the previous iteration. highest score to lowest score. what >are the pros/cons of this method vs. crafty's? > if you don't change your mind during the iteration, the only score you get is for the first move. What do you sort the other moves with when you don't change your mind? >Thank you in advance. > >Sanjiv.
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.