Author: Eugene Nalimov
Date: 17:18:58 09/11/99
Go up one level in this thread
On September 11, 1999 at 20:07:21, Dave Gomboc wrote: >On September 11, 1999 at 19:55:47, Eugene Nalimov wrote: > >>On September 11, 1999 at 16:07:15, Pete R. wrote: >> >>>I know the issue of 6-man tablebases and how much space they would take crops up >>>periodically (and don't throw tomatoes but I don't recall the answer), but is >>>anyone actually generating any, or plan to in the near future? Dr. Nalimov? >>>Just a handful of the most commonly needed would be nice, e.g. krpkrp, or >>>krpkpp. Kasparov vs the World could sorely use kqpkqp. ;) What would it take >>>to generate and store just a few for the most common endings? >> >>I am not doing that. Have no time, and, as I am moving to other group inside MS, >>soon will have no large servers (that can be used over weekend). >> >>I beleive that for 6-man tables you have to use much more efficient backward >>algorithm developed by Stiller (I beleive all others used forward algorithm, as >>it is simpler to implement and debug). Also, 8-10Gb of RAM will be a great help, >>otherwise you'll have to use less efficient disk-oriented algorithms. >> >>And the latter means that you have to use 64-bit CPU as well as 64-bit operating >>system. >> >>Eugene > >Stiller took extensive advantage of symmetries that are not available in >arbitrary six-piece endgames, though. > >Dave Not exactly. He used simple indexing schema, so he could generate moves for the entire 32- or 64-bit vector. So, his algorithm from the beginning was much faster. Of course nothing prevents from generating TB in such a format and then compacting it to (let's say) my indexing schema, but that will be one extra piece of code to write/debug. I beleive Vincent is trying to do something similar, and I suspect he'll find out that TB generation is not much simpler than chess programming. And Stiller used backward algorithm instead of forward one. I.e. he for each "Mate in N" he tried to derive "Mate in N+1"; forward algorithm will check each "unknown" position - maybe it's "Mate in N+1"? So, on each iteration forward algorithm will make much more work than the backward one. Eugene
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.