Author: Ricardo Gibert
Date: 20:00:52 08/28/00
Go up one level in this thread
On August 28, 2000 at 22:36:26, Larry Griffiths wrote: >On August 28, 2000 at 22:21:40, Robert Hyatt wrote: > >>On August 28, 2000 at 19:32:05, Larry Griffiths wrote: >> >>>I have been reading about the History Heuristic and have seen pro's and con's >>>about it. >>> >>>I plan on implementing it to see what happens. This heuristic is related to >>>killer moves and uses the from and to squares in a 64 x 64 array to maintain >>>history information when moves are bestmoves or cutoffs. Each entry has 2 to >>>the depth power added to it when a bestmove or cutoff is found. >>> >>>Would you recommend the History Heuristic, and has anything changed for the >>>better with the method described above? >>> >>>Thanks in advance. >>> >>>Larry. >> >> >>Works fine, but don't use 2^depth... for reasonable search depths, that will >>overflow 32 bit counters almost immediately. I use depth^2 which is much >>safer... >> >>Other than that it works fine. If you don't get a cutoff by the time you have >>tried a few history-ordered moves, you probably should give up and just search >>the rest of the moves in random order. > >I have Jonathan Schaeffer's paper "The History Heuristic and Alpha-Beta Search >Enhancements in Practice". I also figured the counters might overflow and it >looks like he ran his tests to around 9 plys. He also describes that the >history tables can become flooded with information, decreasing their usefulness. > I wondered if this was due to an overflow of his counters at plys 8 and 9. > >Excuse me Bob, but I have not done powers in quite a while and I was thinking >2^depth amounted to shifting the binary value 2 left depth positions. Maybe I >am just tired, but is depth^2 like depth squared? I plan on using 64 bit >counters so I am not worried about overflowing the counters. I thought I would >also try different formula's for calculating the weights. > >Larry. Don't worry, Bob wasn't inventing "new math". The depth^2 function is okay. It is faster to calculate than the function you intend plus it works.
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.