Author: Robert Hyatt
Date: 09:23:09 05/02/01
Go up one level in this thread
On May 02, 2001 at 11:36:08, Ulrich Tuerke wrote: >On May 02, 2001 at 11:21:48, Robert Hyatt wrote: > >>On May 02, 2001 at 10:08:09, JW de Kort wrote: >> >>>dear friends, >>> >>>A few weeks ago i posted a question about cutting my qsearch. With the valuable >>>information you send me i have been able to get a serious reduction for wich i >>>want to thank all to people who helped me realize this. >>> >>>Later i realizes that up till now i have not paid much attention to the move >>>order in the root. The program calculates all the moves and starts the iteration >>>at 1 ply in the order in which the moves are found. If i move is found to be >>>better than all the previous moves tried, this move is placed in front and all >>>the others are shifted one place lower in the order. This gives some kind of >>>ordering after a few iteration. >>> >>>I noticed that mr Hyatt uses a far more advanged methode of ordering but his >>>methode is not clear to me. >>> >>>Also i ask myself if it it worth the effort to try to do a good root move order. >>> >>> >>>I hope some one can help me >>> >>>Thanks in advance >>> >>>Jan Willem >> >> >>Mine is easy to do... and dates back to the early days of Cray Blitz, so it >>has been tested quite extensively... >> >>I save the nodes searched for each root move. For the PV move (the best move >>after iteration N) I set its node count to something larger than any other move, >>then I simply sort on node counts. This puts the PV from iteration N-1 first >>when we do iteration N. If you change your mind at N-1 once, then the original >>best move plus the new best move will have the two highest node counts and they >>will be first. Etc. >> >>Simple and effective. > >This way, you will have heavily extended root moves rather early in your move >list. These moves must not always be good. > >Don't you regard this as a draw-back ? > >The advantage is perhaps that nice tactics (checking sacs ...) will be seen >early. > >Uli I don't see it as a problem. Heavily extending a move that is going to get a quick refutation at ply=2 isn't so expensive in terms of nodes, compared to having to _fully_ search a ply-1 move before you can refute it. The plus here is that it is quite common for a move far down the move list to suddenly take a lot of nodes to refute. At the next iteration it takes even more, until it finally can't be refuted and it replaces the best move. The first time it takes a long time, it moves way up the move list which is good. Ditto for the case where at each new iteration, you flip and flop between two "best" moves. This keeps them both at the top... and will include that "soon-to-be-best move as well...
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.