Author: Robert Hyatt
Date: 14:22:54 08/16/99
Go up one level in this thread
On August 16, 1999 at 15:36:50, Paul Tillotson wrote: >On August 16, 1999 at 15:16:30, Robert Hyatt wrote: > >>On August 16, 1999 at 15:07:59, Paul Tillotson wrote: >> >>>On August 16, 1999 at 14:00:22, Robert Hyatt wrote: >>> >>>>On August 15, 1999 at 21:16:42, Randall Shane wrote: >>>> >>>>>During my alpha-beta search, if I run out of time before I complete a iteration, >>>>>is it safe to just go with the move from the previous iteration?? >>>>> >>>>>What other strategies are used? >>>>> >>>>>Thanks! >>>> >>>>(a) use the best move backed up in _this_ iteration (even though you have >>>>not completed it) unless the score has dropped markedly from the best score >>>>the last iteration. In that cause, use more time. >>>> >>>>(b) use the move from the previous iteration if you haven't gotten a best move >>>>for the current iteration yet, unless you have failed low in the current >>>>iteration. In that case, use more time. >>> >>>if (a) then what would be a markedly dropped score? >>>Would .5 pawn or a whole pawn be in the ball park? >>> >>>Regards, >>>Paul >> >> >>Actually I have a staircase of values. I am quoting from memory (you can check >>time.c in crafty for actual test)... >> >>if the score drops 1/3 of a pawn, I use 2x the target time to try to 'get that >>positional score back.'. If I am dropping a pawn, I think I will use up to 5x >>the base target time. In cray blitz we were more dynamic than that, but I am >>not sure it was a good idea (ie positional drop = 1.5x, pawn = 2x, piece = 5x, >>mated= use almost all remaining time to find a way out...) > > >Seems very reasonable. One other directly related question. >If your score drops a full pawn then you give it 2x. Yet you found the >positional score at about 1.1x (ie just needed a little more time) then is >it worth stopping the search there? Or just finish the rest of the 0.9x time? > >Regards, >Paul I'm not sure what is best here. What I do is about once per second, I ask the question "is the root score > last_root_score-30"? If the answer is yes, I time-out and quit. If the answer is no, I continue until the 2x time limit has expired. If any move returns a score to the root that will break the above test, I stop instantly (IE on the first move that gets within .30 of the previous search.) It might be better to search more to see if there is a better move that will get even more of that score back...
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.