Author: William Bryant
Date: 14:03:44 03/01/00
Go up one level in this thread
On March 01, 2000 at 15:55:47, John Coffey wrote: > >> >>The idea is to generate the moves as late in the cycle as possible to avoid >>doing the work if you generate a cutoff sooner. >> >>william >>wbryant@ix.netcom.com > >I envisiioned that one would have to generate all the moves first to determine >which moves are captures. You say to look at captures and killers first. How >do you know which ones are killers? (What is a killer anyway?) Doesn't this >require that you generate a complete move list first? Break down your move generator into two functions, one that generates captures and promotions, and one that generates only non-captures. Hopefully, one of the captures will generate a cutoff and the non-captures will not have to be generated at all. Killers are moves that are non-capture moves that are saved at each ply. If a move is good enough to reset alpha at one part of the search, it might be good at another part of the search at the same ply. Because winning captures are searched before killers, exclude capture moves from a killer list. These moves are usually kept on a special list for each ply. > >I figure that bitboards are probably used to generate captures somehow, but then >later when you are looking at the remaining legal moves, how do you check to see >that you haven't looked at a move already? Since your are searching the hash/pv move first, you need to flag this move as searched when you generate you move lists. The same applies to the killer moves. You need to flag them as searched after generating the rest of the move list. >At some point you have to sort the >moves? If so then what method do you us to sort the moves? I use a simple sort (not sure of the techinical term here). I get the next move from the move list and search it. If it does't generate a cuttoff, then I get the next move, etc... Other than the root list, sorting the entire list may not be necessary if on of the initial moves generates a cutoff. william wbryant@ix.netcom.com >John
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.