Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: A question of time

Author: David Dory

Date: 04:08:59 01/24/02

Go up one level in this thread


On January 24, 2002 at 04:47:30, guy haworth wrote:

>Would welcome sources of information on how chess-engines budget their time.
>
>Are the heuristics out there more 'conservative' than human players?  What
>criteria are used for departing from some norm-budget/move, assuming there is
>one?
Time management is always critical for chess programs. Iterative deepening is a
term used to describe the process most used to integrate time management into
the search of the program.

Basically, the program figures out the average time per move and uses that until
near the end of the time control. It explores the moves to the first ply (a move
by one side), and checks how the time is now doing, periodically. It should
still have plenty at this point, so it saves it's best move found so far, and
goes to the next ply down in the search tree, and repeats, again saving it's
data for best move at this depth.

This is repeated until the time used is about 75% or so of the average time per
move which it has figured out. If it has found a good move (and that definition
varies from program to program), at that point it will then play it. If it
hasn't, rather than play a bad move, the program will keep going deeper into the
search, again, saving it's info as it goes (just in case the operator presses
the key telling it to "MOVE NOW!"). When 100% of the average time per move has
been reached, the program will play it's best move unless it knows it's still a
bad move (bad might be defined as a move which lowers it's score by more than
half a pawn). Again, the program spends more time searching for a good move, but
at some point, say 150% - 200% of that average time per move, it will make the
best move it has found, regardless.

This is all based on the idea that some moves (such as book moves), and simple
recaptures will use almost no time, while other positions need far more time to
search before a good move can be found.

>
>Do chess engines modify their:
>
>  a)  opening books,
>  b)  search and/or evaluation methods
>
>according to the time regime for the game?

Not any program of mine! :-)

>Are chess-engines able to handle the increasing variety of time-regimes being
>defined now?

Yes, using iterative deepening, time management is not the "*&^%/!!" (insert
favorite cuss words here) problem it used to be. At the end of time controls it
get's a little tricky.

>What time-regimes are there and is there an established notation
>for defining them?
I would call it time regimens or time controls, but not "time-regimes".

Why not log onto the chess servers and see?

(Game in 'n' minutes, 'n' moves in 'n' minutes (primary), plus 'n' moves in 'n'
minutes thereafter (secondary), and Fisher come to mind.)

Dave




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.