Author: Vincent Diepeveen
Date: 04:43:28 07/22/99
Go up one level in this thread
On July 21, 1999 at 15:00:43, Robert Hyatt wrote: >On July 20, 1999 at 19:51:57, Vincent Diepeveen wrote: > >>On July 20, 1999 at 17:30:54, Dann Corbit wrote: >> >>>On July 20, 1999 at 17:26:49, Dave Gomboc wrote: >>>[snip] >>>>None of the commercials I know of provide source code, so this might be tough to >>>>arrange. Is your opinion that Crafty would be adequate? >>>I see possible solutions: >>>0. We are provided with an object module and permission to use it for the test >>>only. >>>1. We keep a record of the fastest currently tested implementation. >>>2. We hold a contest with prizes for the winning algorithm. >> >>I nominate the nullmove, >>althoug of course alfabeta logically should qualify first, >>alfabetapruning isn't very hard to invent. Nullmove is though. >> >>First prize for that nomination would be: >> "best algorithmic improvement of the century" >> >>Nullmove has been nominated as #1 by me. >>Alfabeta as second. >>YBW as third. >>endgametablebases as 4th. > > >I would disagree on #1, because without alpha/beta, null-move doesn't work at >all. So I'd go with alpha/beta, then null-move. I have to disagree that alfabeta is more important. Alfabeta is rather simple to invent. Only negamax makes it more complicated. >YBW isn't a particular algorithm. It is just a noticable fact produced by the >way alpha/beta searches a tree, so it is a basic part. I might go along with >the concept of 'parallel search' instead, as this will become more and more >important is time passes. It's a great improvement nevertheless. Whether it's a logical concept, that's discussable. I have to agree, but there are no other things in parallel search that are tough to discover. Tricks like i do, like never splitting more than once in a position (maximum of 2 processors at 1 position; don't do this with too many processors!), not being parallel but completely recursive; i consider all these small improvements, which alltogether give great performance, but they're not worth nominating. >Probably even more important is the various 'extension' ideas developed over >the years... out of check, one-reply, recapture, passed pawn push, singular >extensions (and various derivatives), threat extensions, etc.... I disagree on that. It's dead simple extensions. Selective search hasn't been written about in literature yet, i wonder why, it gives my diep at least a 100 points extra in blitz. Extensions are simply easy to invent, no matter how you call them. You can use 'units' for a ply, and less units for an extended move, whatever you want to do. Extensions are not giving that much compared to nullmove and a good qsearch, except for check extension. However check extensions are *so logical* and can be done in so many ways that it's tough to nominate them. At slower hardware than we have nowadays extensions are much more important than nowadays. The only thing i wonder about is how Genius is doing its tactical search. Any ideas here? I mean it seems searching fullwidth, but it can't search fullwidth as it needs too little nodes to get to a plydepth, yet it sees tactics very deep. It seems to extend in a smart way near the leafs?
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.