Author: Angrim
Date: 17:37:27 01/30/02
Go up one level in this thread
On January 30, 2002 at 14:41:52, Dann Corbit wrote: >They aren't really tablebase files anyway. They are approximations to them >only. They don't give a best move. Instead, they give a winning move or a >drawing move. But not necessarily the shortest one. They would be (depending >on the type and number of pieces) much smaller than Nalimov tablebase files. > >You don't believe that, which is fine. But it's a fact. I wish you would stop posting this over and over. In this and quite a few other posts you have stated that this method would be a huge improvement over current tablebase formats, while clearly not understanding current tablebase formats. quick summary of the state of the art in tablebase storeage: 1. all safe symetries are used to reduce the number of positions. for tables with pawns, this is left/right and side to move. for tables without pawns this also includes top/bottom and diagonal. 1a. positions in which the two kings are touching are also excluded. 2. each position is stored as a combination of filename(KBvkr or such) and an index into the file, no bytes are used to store the position. At the index into the file coresponding to a unique position, the value for that position is stored. 3. the resulting table is then split into 8k blocks, and compressed with a block compression algorithm optimized for decompression speed. The resulting compression for pawnless tables is roughly: 20x reduction in number of positions from the symetries and avoiding illegal positions. 6x compression from the block compressor so 120 position's values stored per byte of tablebase file. for tables with pawns, it is: a little over 4x from the symetries and avoiding illegal positions. 6x compression from the block compressor so 24+ position's values stored per byte of tablebase file. And note that these are averages for all 5pc endgames, not just the best case situation with only sliding pieces. Angrim
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.