Author: Robert Hyatt
Date: 10:49:55 10/29/99
Go up one level in this thread
On October 29, 1999 at 00:25:42, Michael de la Maza wrote: >Imagine that you are writing a naive time allocation algorithm for a chess >program that will be playing against humans. Suppose that the time allocation >can only be a function of the move number. > >To simplify things, assume that the first 10 moves are all book moves and are >played instantly. Also assume that the computer has a total of 60 minutes to >make all of its moves. > >Here are some possible time allocation algorithms: > >1. Allocate 30s per move >This would allow the machine to make 130 moves (the 10 instant book moves + 120 >more) which is enough to complete all but the longest games. Good humans might smoke you here. You either beat them tactically, or they will lock the position and watch your burn your clock down to zero. They only have to push a pawn every 50 moves while waiting. > >2. Allocate 60s for moves 11-40 and 20s per move thereafter >With this algorithm, the machine could make a total of 100 moves, and would >spend more time immediately following the opening and in the middle game. > >3. Allocate 120s for moves 11-39 and 2s per move thereafter >The idea here is to win during the middle game. > >What would you do and why? None of the above. What about "target=remaining_time/C" where C is a number you try and like (say 30). This means that (a) you use more time early in the game where it is important; (b) you will _never_ run out of time as you only used 1/30th of your remaining time for the next move. That's about as simple as you can get. > >More generally, if we assume that the strength of the program is tactics should >the time allocation algorithm allocate more time to the middle game to support >this strength or should it allocate more time to the end game to cover this >weakness? Should the program allocate a lot of time in the middle game in the >hope of getting a decisive advantage or should the time be spread out more >evenly to avoid blunders? > >Remember that the program is playing against humans, not computers. > >Michael
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.