Author: Andrew Williams
Date: 16:10:44 02/15/00
Go up one level in this thread
On February 15, 2000 at 18:10:12, Sanjiv Karnataki wrote: >Hi All, > >I am at the point where I am about to put time controls in my program and >thought I will ask the experts before trying to do it all. > >a) determine the time available for the next move. > >b) While time is unexpired, search. At expiry of time use the best move from the >last iteration. If the best move is a negative evaluation increase the time by >some factor. > >That is the easy part. > >a) How do you determine if you don't need to use all the time ? eg: after >iteration 10 the score is +20, would you save time by not looking further? > I only do this sort of thing if there's only one move available, in which case I do a very short search (4 ply I think) so that I can be sure I've got a move to ponder on. Otherwise I just search for the full allocation. This time isn't exactly wasted; it fills up the hash table with (hopefully) useful info. Have you got access to old issues of ICCA Journal? If you do, you might want to look at Chrilly Donninger's article "A la Recherche du Temps Perdu: That was easy" in vol 17 No 1 March 1994. I don't know if this is out of date, but it discusses the issue. >b) in case the time control is 40 moves in 120 minutes, it gives 3 minutes per >move. how do you determine what time control to use if the games is all moves in >xx minutes? How do you handle increments ? > The increments thing is something I've been meaning to ask here for a while. For your other question, I simply divide the time available by 30. Once we reach the endgame, I reduce this to 20. With an increment, PostModernist adds the increment to the time after the division. Before CCT1, I noticed that PM would use less time than other programs in increment games. So where there's an increment, I reduce the divisor somewhat (maybe 28 and 18, IIRC). I can't help feeling that this isn't ideal, but I couldn't think of anything better at the time. >Anything else that I should watch out for? > >All help is appreciated as always, and yes I will do the moral and proper thing >by looking at what crafty does...but it is a complicated program and sometimes I >just find it easier to ask the question here ... > >Thank you ! > >Sanjiv. Cheers 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.