Author: Ren Wu
Date: 13:51:19 12/04/01
Go up one level in this thread
On December 04, 2001 at 10:01:23, guy haworth wrote: >It is clear that the EGT-generation method of Wu/Beal predates the April 2001 >publication by UK. It seems that the WB-approach has been in the pipeline since >work at QMWC (London) started about 1994. That's right. >It has been published in the International Systems Journal (2000) and in the >ICGA_J_v24.3 (Sept 2001). > > >For the record, you should note that the Wu/Beal method has two minor flaws, >easily spotted when used: No, this is a wrong statement, see below. >a) accidentally omits to inherit subgame-wins-in 0 and 1 in DTC-mode > ... although this was clearly intended No. All subgame positions will be initialized. This was done in DoInitialize() function in our algorithm. > >b) does not ensure that there are at least as many cycles as the max depth of a >subgame win. [ It is theoretically possible for the last cycle to have >discovered nothing new but there still to be subgame wins to inherit. ] In fact, our algorithm handle this nicely. And that is exaclty the reason we use to funtions, DoneWhite() and DoneBlack() as the terminate condistions, instead using the number of win/lost positions in last iteration. >Also, I suspect the Wu/Beal method does not make full use of bitmaps, minimising >the use of the depth-databases as much as possible (as per Ken Thompson, 1986). >John Tamplin proposes bitmaps for 'residual unknowns' as well as 'positions >resolved so far'. Anything that improves efficiency at 6-man level is good >news. >So there are improvements to be made, even now. It should be clear that the main focus of the paper is to present a CORRECT algorithm that required MINIMAL amount of RAM to build databases with FULL information. Our algorithm will benifit from more available memory too. And there are many ways to utilize these extra memory. For your record, we just finished a chinese chess endgame databsse with about 5.8 billion positions per side, or 11.6B total, in 77 hours on a Pentium 3 machine, with 1 GB of memory. I am pretty sure that 6 -men chess endgame database can be built within same amount of time on same machine. Ren.
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.