Author: Les Fernandez
Date: 10:28:47 01/15/03
Go up one level in this thread
On January 15, 2003 at 12:49:45, Marc Bourzutschky wrote: >On January 15, 2003 at 11:40:29, Les Fernandez wrote: > >>On January 15, 2003 at 11:33:16, Marc Bourzutschky wrote: >> >>>On January 15, 2003 at 09:48:25, Robert Hyatt wrote: >>> >>><<snip>> >>>>>If you are really desparate for those 5-1 endings (and like Eugene I don't see >>>>>the point of generating them) you can use Johan de Koning's FEG program >>>>>(available on www.chessmaster.com) to do it. For 5-1 endings you'll need about >>>>>300MB of RAM and a few gig disk space. A typical pawnless ending will take >>>>>about a day, an ending with pawns about 3 days, on a 2 Ghz machine. >>>>> >>>>>-Marc >>>> >>>> >>>>Does it produce DTM tables that will work with Eugene's probe code? >>>> >>>>Or is this the win/lose/draw stuff? >>> >>>It has full DTM information, but stored in a quite different format from >>>Eugene's, and therefore not accessible by his probe code. It should be possible >>>to translate one format to the other, but such a utility does not yet exist. >> >>Hi Marc, >> >>Listen when you refer to a different format then Eugenes are you saying that in >>it exists wtm draw,bwtm win, wtm mate, wtm broken, btm draw, btm win, btm mate, >>btm broken info? I am just trying to understand it because perhaps I can write >>a converter to go to the format I need if the info is there. Just a thought >> >>Les > >The format refers to the binary containing the actual data, and since the format >is not public it would be almost impossible to write a converter yourself. > >During generation, the program also produces a stats file, similar to Eugene's >.tbs file. The stats file contains for each DTM the number of positions with >that DTM, i.e., you get complete wtm win and btm lost info (for 5-1 there >clearly are no wtm lost or btm win positions). Broken positions and draws are >not separated in the stats file, but it is easy enough to write a program to >count broken positions, and then you can infer the number of draws by >subtraction. > >One difference between the stats file and the .tbs file is the treatment of >symmetry for pawnless endings, so you can't compare counts directly. The .tbs >file restricts the two kings to 462 squares, while in the stats file it >restricts one of the kings to 10 squares, so you will consistently get more >positions. The compressed version of the chessmaster data actually applies full >diagonal symmetery to the data, so you get even fewer positions than you get in >the .tbs file, which only has a partial implementation of diagonal symmetry. >But since the stats file is produced during generation rather than after >compression it contains more positions. It is actually possible to collect >win-loss statistics on the compressed files as well (ply by ply), which will >give you the stats with fully implemented symmetries. For example, consider >this wtm and mate in 2 position in KRK: Thanks Marc for an excellent description. It appears to boil down to comaparing apples and oranges as far as I can see. I certainly do not want to introduce any variablitiy to what I am looking at! Appears to be enough variability in both approaches to concern me. Hmmm guess I will have to think further on this one. Les > >[D] 8/7R/8/8/8/2K5/8/k7 w > >the compressed chessmaster data would only count this position once, while in >the .tbs file you would get a count of 2, because it also contains the position >obtained by reflecting along the a1-h8 diagonal: > >[D] 6R1/8/8/8/8/2K5/8/k7 w > >-Marc > >>> >>>The key benefit of Johan's program is the small footprint. It only requires >>>enough RAM to store the side to move, and as a result only needs about 15MB for >>>6-man endings (excepting the pathological 5-1, which requires 300MB), regardless >>>whether pawns are present or not. The program is also significantly faster than >>>the publically available version of Eugene's tbgen program (although it seems >>>Eugene has a faster version now). The complete 5-man set takes about 2 3/4 days >>>to build on an AMD2600.
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.