Author: Dave Gomboc
Date: 10:53:07 04/14/99
Go up one level in this thread
On April 14, 1999 at 11:42:31, Keith Kitson wrote: >It occurs to me that our abilities to build tactical awareness into programs has >been quite successful. Our major aim now should be to build in better strategic >elements. Better planning and long term follow through of a plan. I know >strong chess players, see patterns, and look at the board dynamically. If we >can build this sort of awareness into the programs, without detrimentally >affecting the tactical side then I see some strides forward being made. How >long it will take to achieve this I find it difficult to predict. One of the >itme that i find unusual, is that for some programs that use Hash Tables they >clear the hash tables down after every move. That is not the way a human >players memory works. They define a plan, perhaps with some alternatives, then >they use they use the plan over the course of the game. They don't start again >from scratch to build a plan with each ply (or half move) that is made. I >reckon some strides forward could be made here in time savings, to allow deeper >lines of thought within the given time controls. there again the deeper the >program calculates the less the accuracy of the line due to search curtailment >criteria. I shall watch with interest over the next few years to see how things >develop. > >Kkitson I think that it is a mistake to attempt to model chess software based on "how humans play". The von Neumann-style CPU has different strengths and weaknesses than our own brains. They are relatively bad at pattern recognition, and relative good at search, so it behooves us to harness the power of its ability to search, and indeed, that is what every successful program is doing. With respect to the clearing of hash tables, the number of leaf nodes dominates the number of internal nodes in a position, so it's not extremely important to keep it. Some developers find that the increase in strength in storing information in the hash table that is not valid from one search to the next is worth more than the (small) decrease that occurs by having "cleared" (and it is probably not really cleared, just timestamped) the hash table. Dave Gomboc
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.