Author: Robert Hyatt
Date: 07:42:05 07/22/04
Go up one level in this thread
On July 22, 2004 at 00:16:36, Eric Oldre wrote: >On July 21, 2004 at 23:13:26, Robert Hyatt wrote: > >>On July 21, 2004 at 23:03:59, Eric Oldre wrote: >> >>>On July 21, 2004 at 22:09:58, Robert Hyatt wrote: >>> >>>>On July 21, 2004 at 16:55:07, Eric Oldre wrote: >>>> >>>>>lets say you are searching a position with iterative deepening. >>>>>at depths 3,4,5 it finds a best move of Qd3. then at depth 6 >>>>>the best move changes to c4. >>>>> >>>>>When you start searching to depth 7, you clearly would want the >>>>>first move you search to be c4, and this shouldn't be a problem >>>>>if your transposition tables are working right. Since the best >>>>>move should be stored in it. >>>>> >>>>>however, a good move to try 2nd would be Qd3, but your transposition >>>>>tables aren't going to let you know that, at least not in the implementations >>>>>that I have normally heard about. >>>>> >>>>>I was thinking that one way to overcome this would be to keep 2 "best" moves >>>>>in the trans table. one to try first, and one that was a best move, before >>>>>being replaced by a deeper search. >>>>> >>>>>Are other people doing this? or have you discovered that the cost is greater >>>>>than the benefit? >>>>> >>>>>Eric >>>> >>>> >>>>Nobody uses the transposition table to order the moves at ply=1. >>> >>>Well not quite, I do! >> >>You only have _one_ position at the root of the search. How many moves can you >>store in one position? :) >> >>More than one is a huge waste since "best" is all that counts in alpha/beta... > > >i think that you might have misunderstood what i was trying to say, or i >misunderstood you... > >I meant i get the first move to search from the trans table. and then use SEE >after that. > > OK. That was not what it sounded like. First move from hash is fine. But sorting the remainder of the moves with SEE/etc is not as good as other ideas, such as when you find a new best move at the root, keep the previous best move as #2 as in many positions the two moves flip-flop each iteration... > > > > > > >>>Of course i hope to get that replaced in the future with something better. :) >>> >>> >>>> Much easier is >>>>to just generate the ply=1 move list before you start the search, and as you >>>>find a new best move, move it to the top of the list, pushing others down 1 >>>>slot. This makes you search the best move from the previous iteration first, >>>>then the next-best move from the previous iteration second, etc...
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.