Author: Uri Blass
Date: 08:25:39 10/27/01
Go up one level in this thread
On October 27, 2001 at 10:28:08, Robert Hyatt wrote: >On October 27, 2001 at 06:42:09, Ron Langeveld wrote: > >>I was wondering whether you have generated the 6 piece EGTB yourself, and if so >>what your experience is with such a task. > > >I generated all the 3-4-5 piece files once. I have not tried a 6 piece >file as they require a lot of memory and they won't generate under Linux >(or at least they would not under older kernels because of a 32 bit file >pointer limit). > >Eugene is doing the generating on alpha-based machines. They represent a >huge committment of computing cycles. > >> >>If more tables could be generated, how to go about ? Maybe more people of this >>forum can participate in a cooperative effort, but then of course they would >>want to know how much time generating just one table could take given a certain >>hardware configuration. There might even be some practical limitation with >>regard to hardware. I just don't know. Maybe you do. > > >Eugene is working on the generator/probe code. He has already generated some >6-piece files that the current probe code won't handle properly because they >are well beyond 2.0 gigabytes in size. He has said that this will be fixed >at some point... I think that it may be productive to remember tablebases of only win loss draw and when you get the relevant position the program may connect to some site in the internet in order to read more information about the right move. There is a known problem of the 50 rule move but this problem happens because the tablebases are wrong and the only way to solve it with no mistake is simply to generate the right tablebases for the 50 move rule that are smaller. Another possible idea is to use the original tablebases and get an estimate to the distance to conversion in every position by playing games between the program and itself. The games can be of only 1 ply if you start from the position that are closer to mate because after 1 ply if it is not a capture or pawn move you know an estimate for the distance to conversion based on a previous game. It is not exactly correct to do it but I think that you need less time to do it than generating the right tablebases and it is better than using the original tablebases. Uri
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.