Author: Robert Hyatt
Date: 18:01:56 08/21/01
Go up one level in this thread
On August 21, 2001 at 16:58:30, Scott Gasch wrote: >Hi, > >I am thinking about time management these days. Currently I have rules that >make the engine take extra time when it's just out of book or if it sees two >fail lows during a search. It will also move faster if the root position is >blocked and it's considering a move that won't help unblock it. > >Additionally it moves faster if it has a clear recapture (yes, I know this is >dangerous but I have several safeguards in place and I have not been bitten by >it yet). > >I have seen other engines that handle time differently, though. For example, I >don't take extra time to resolve a fail high at the root. And with the "take >extra time and resolve the fail low" rule, I only take a little extra time and >will play without resolving if that time runs out. I have observed ferret >seeming to take a bunch of extra time in these cases though. Should we always >resolve root fail highs or fail lows? How important is this? > >Thanks, >Scott I don't resolve fail highs at the root either if I run out of time. I don't see the point. But you did miss one important idea that has been around forever: don't time out the search when partially thru analyzing a move at the root. And I mean "any" move. IE once you start searching a move at the root (unless it is the first move on a new iteration) and you run out of time, continue searching until you finish this move. This is important. Because often on the last iteration you will start a search on a non-first move, and if you have enough time, you will change to that new best move. But only if you have enough time. Make sure you do by not stopping until you search that move completely. With null-move most of the time this happens instantly anyway, so it costs you nothing. Occasionally you will discover that you can not prune away things you could in previous iterations,b ecause this move is about to become a new best move. Why stop until you know it is or it isn't??? easy to implement...
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.