Author: Gerd Isenberg
Date: 13:24:36 07/26/04
Go up one level in this thread
On July 26, 2004 at 08:25:45, Ricardo Gibert wrote: >On July 26, 2004 at 07:39:42, Bas Hamstra wrote: > >>>>I basically try to calculate number of expected moves to finish the game or >>next time control and use this information to decide about target time and I >>>>also use the time of the last iteration to decide if to do another iteration >>>>usually try to finish iteration in most cases with the idea that after >>>> finishing >>>>an iteration there is a smaller chance to change my mind when the main >>>> exception >>>>are cases when the program pondered enough time). >>> >>>Bad idea. Start the next iteration even if you don't think you will have time >>>to finish it. You might fail low. Wouldn't that be nice to know? :) >> >>On the other hand there, have you ever summed up the total time that is wasted >>at the first move? Suppose your target time is 60 seconds and you have just >>finished an iteration, which took 50 seconds. There won't be enough time to >>calculate the first move of the next iteration, so you might as well save 10 >>seconds. >> >>Bas. > >The natural moment to stop and move is to stop at the end of the iteration, but >this does not seem to really be the best moment. Usually when you fail low, it >is near the beginning of the next iteration. If you begin the next iteration and >do not fail low in the first few seconds, then you can be more confident in the >efficacy of the intended move. If you fail low instead, then that is really nice >to know, yes? Those 10 seconds will not be wasted. > >This is not my idea. I learned this from >http://brick.bitpit.net/~marcelk/2002/marcelk-thesis.pdf from page 17. It's an >astute trick. Yes, i didn't try it yet but i will consider it. "One Jump Ahead" iirc, Jonathan Schaeffer did similar in Chinook. To distinguish between best so far "sound" and "unsound" or "risky" moves is IMHO also worth to consider if asking for a next iteration. The question is what is sound and what is unsound and risky, but at the root one may use some rather huge move evaluation. Same is true if moves are probably conform with some attacking or defending plan or follow the previous move(s) in some locigal way, let say plan is minory attack on the queen side and to play b4 after Rb1 or so. Otherwise, for what reasons ever (dumb eval?), if the program likes to castle kingside, where the opponent has already a half open file for his rooks... With some more knowledge at least at the root it may be considered unsound to force the search going deeper - probably seeing the problems coming next ply and finding a better move. Own "Nullmoves" in the PV are probably an indicator, that the search tends to delay some local maximum to the leaves. May be also a hint to search further? That often occurs after winning some material but the opponent has increasing compensation and/or threats to get material back. If the best move so far is somehow committing, most often pawn moves, but we had an "better" one/two iterations before, one may based on score, take more time too, hoping for a more "flexible" move! Of course one should study odd/even behaviour in quite positions, to get an idea whether it is worth (flipping rate?). Gerd
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.