Author: Ricardo Gibert
Date: 15:38:13 07/20/99
Go up one level in this thread
On July 20, 1999 at 15:07:24, Bruce Moreland wrote: > >On July 20, 1999 at 12:39:48, Dann Corbit wrote: > >>We might be able to achieve something like that by looking at certain >>implementations. For instance, we could find out the record for legal move >>generation for a normalized number of CPU mips. After finding out what the >>fastest routines are, we could see what algorithms they use. We could find >>programs where they have replaceable drivers for search techniques and have >>benchmarked both techniques, etc. > >If I understand correctly, what you want to do is try to determine which data >structures are best for chess. > >This may be possible but it would be very difficult for several reasons. > >The code surrounding the data structures can be written in many different ways, >which may have different characteristics on different architectures. > >Different implementations may work better in different phases of the game, and >if you want to try to figure out the ramifications of that, you're going to have >to talk about incredibly esoteric and resource consumptive experiments. I don't >have time and energy for that, personally. > >Individual elements of a chess program do not exist in isolation, there is a lot >of interaction with the other elements. A fast move generator may generate >moves really fast, but it doesn't do you much good if the makemove is slow, or >if there is some eval term you want but you can't get efficiently because you >can't incrementally calculate it, or it's so large that it blows instruction >cache when you actually write a chess program around it, etc. > >bruce Board representation, search algorithm, evaluation routine,...all interact with one another in ways that make isolating one component of a program from another problematical. How you generate moves affects how you perform (and how quickly) evaluations. How you search effects how you should eval - coarsness of eval for example. Idea is well motivated, but its too hard to do this "scientifically". You may generate a lot of data, but what will it mean?
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.