Author: Robert Hyatt
Date: 11:46:59 03/19/02
Go up one level in this thread
On March 19, 2002 at 13:40:54, bob o wrote: >1. I think I read something (I could be wrong) where they said that someone made >a program so that some 6-man tablebases could be made in chunks, so that a bunch >of different programs could split up the task of making these gigantic files, >kinda like those guys that are looking for aliens. Does anyone know anything >about that, or was my source incorrect? > >2. For Prof. Hyatt: On the ftp site, there are some tbs's for tablebases that >are not available on the site; I think that they are all 5+1 files, I could be >wrong... is it possible to obtain the TB files somehow? Eugene is busy building more files (3 vs 3 I think). Unfortunately some of them are so large that a 32 bit address space won't work. Eugene is going to modify his probe code to fix this, but at present it is not ready. For that reason, those specific 6 piece files are not yet available for download as they would cause significant problems if they were used.. > >3. When playing around with tbcomp.exe I noticed that if you decompress the >downloaded files and then compress them again with large values for the size of >the encode block size (the -e parameter) you can change (and reduce) the size of >the files; for the 6-man files I tried so far I got very significant size >savings, but I was not able to run them under crafty (I think it said something >like illegal block size or something, it's been a while... I can try again to >reproduce this if requested). Is it possible to compress them in this manner to >get savings in file size? And if it is possible, what are the chess-playing >ramifications of this? The bigger the blocksize, the better the compression. But the slower the access because an _entire_ block must be read in and decompressed, to access any value within that block. The current 8K blocksize was chosen after a lot of testing... Making it bigger will reduce the file size and slow down the probe code at the same time. > >4. Is it possible for a program to need just one of the tablebase files instead >of the pair? For instance, let's say that you had an endgame of: white- ke1 pawn >e2, black- ke8. And suppose you only had kpk.nbw.emd. Is it possible for the >program with white to say, "Ok I know that I have a mate in N moves." Then it >makes a move. Then black says "I can't find the correct TB file. But I found >kpk.nbw.emd. Now let me search every possible black move for me and white's best >reply, and leave him in the worst possible position." It seems that if this is >possible, then you could cut the storage space in half by doing a quick 2-ply >search. That is possible in some cases. The problem is that there are identical positions with black and white on move. If you don't have one, then you can go to the computational task of inverting the board, which inverts the side on move, but that takes a bit of time... > >5. I'm not quite clear on what info is in which TB file. For instance, let's say >we have a series of kqkr endings. In which files are the following (kqkr.nbb or >kqkr.nbw) > 1. white's queen vs. black's rook, white to move > 2. white's queen vs. black's rook, black to move > 3. white's rook vs. black's queen, white to move > 4. white's rook vs. black's queen, black to move all 4 pieces, white to move positions are in .nbw files, black to move positions are in .nbb files > >Sorry if these questions were answered here before, I am a rather recent >newcomer here, and I do not have a great understanding of TB files, but am >fascinated by them. Thanks in advance. > >Bob
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.