Author: Robert Hyatt
Date: 06:05:25 09/02/00
Go up one level in this thread
On September 01, 2000 at 10:21:13, Dave Gomboc wrote: >On August 31, 2000 at 13:36:54, Bruce Moreland wrote: > >>On August 30, 2000 at 05:41:01, Severi Salminen wrote: >> >>>>>>Why do you add a value depending on depth (2^depth)? Why not just increment by >>>>>>1? Just asking because I'm new to chess programming techniques and I'm starting >>>>>>to program my own creature... >>>>>> >>>>>>Severi >>>>> >>>>>I believe the idea was to give higher weights to nodes near the root since they >>>>>are not updated as often. >>>>> >>>> >>>>and they are also more important, as they are *still* good with a deeper search >>>>tree below them. >>> >>>Oh, I got it. I thought it was the depth in which the cut off was found, not the >>>depth remaining below that node... >>> >>>So is this what basically happens: >>>1. you generate pseudo-moves >>>2. you give captures a big priority plus >>>3. you add the corresponding history value from history[from][to] to priority >>>value >>>4. make the best move >>>5. inc history value in array if cutoff found (or fail high) >>>6. after search decrease history values a bit >>> >>>Right? >>> >>>Severi >> >>The 2^depth thing is based upon the belief that chess programs play better if >>they use "cool" math like "^" rather than "boring" math like "+". >> >>bruce > >No, it's based on wanting to give higher increments to moves when they succeed >near the root of the search than when they succeed near the tips of the search. >An absolute shift left operation (<< ; how 2^depth is implemented) both gives >this effect and executes extremely quickly. > >As an aside, the history heuristic is something that was created 15 years ago, >and it wouldn't terribly surprise me if programs have changed enough over that >time that it's no longer valuable. I believe it's still a net win for Jonathan >in his Chinook program, though. > >Dave Remember that the history heuristic is just a more general case of the killer move idea, which is over 25 years old. And so far as I know, either still works very well in today's chess programs. If you already have killers, history might not help a lot, but I haven't seen a case where it hurts, once it is implemented correctly...
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.