Author: José Carlos
Date: 06:20:43 11/22/00
Go up one level in this thread
On November 21, 2000 at 21:46:07, James Swafford wrote: >On November 21, 2000 at 10:23:04, Brian Richardson wrote: > >>On November 21, 2000 at 09:00:16, Rafael Andrist wrote: >> >>>Normally, quiescence search is used at the end of a tree. It seems to me, that >>>it makes only sense to do qsearch after the last increment. But if there is a >>>time limit and after the last increment is no time to do a qsearch (or you must >>>stop in the middle of the search), the program can have a completly wrong >>>evaluation. How is this handled in chess programs? >>> >>>Rafael B. Andrist >> >>Most programs use iterative deepening, so the principal variation (and a best >>move so far) is always known from the prior (last) completed iteration. If the >>qsearch or the "full width" search does not complete before a time limit I think >>the results of that incomplete iteration are generally discarded. For example, >>the search (and qsearch of each leaf node) finishes for iteration at depth n. >>Depth n+1 does not finish. The best move from the depth n search is used. > > >I don't think you need to complete the iteration, just the move. >Let's say at iteration n your best move was m1. During iteration n+1, >move m1 is the first root moved searched and gets a score. Then >m2 is searched (and is finished) and gets a score s2. Let's say the >time expires while searching m3 and you abort the search - so only >two moves were completely searched in iteration n+1. >If s2 > s1, I'd say choose move 2. > >Maybe that's what you meant; I'm not sure. :-) > >-- >James I think we all do this. Even if you stop the search after a fail high without resolving it (getting the exact value), you accept that as the best move, and play it. I don't stop the search in the middle of the tree in Averno. I only stop in the root, but indeed I don't need to complete all the root moves for a given iteration. José C.
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.