Author: Dann Corbit
Date: 00:30:44 04/17/04
Go up one level in this thread
On April 16, 2004 at 17:03:47, Dann Corbit wrote: >Considering files like this one: >ftp://ftp.cis.uab.edu/pub/hyatt/TB/tbs/kqqqkr.tbs > >I think we can make a more compact bitbase representation. > >The only really interesting positions are where the dominant side does not >dominate. > >From the above file, that boils down to this set: > >wtm: Lost in 0: 14415 >wtm: Lost in 1: 5898 >wtm: Lost in 2: 3397 >wtm: Lost in 3: 1730 > >wtm: Draws: 52698 > >btm: Draws: 495121 > >btm: Mate in 4: 12110 >btm: Mate in 3: 31861 >btm: Mate in 2: 53435 >btm: Mate in 1: 123889 > >A total of 794,554 positions in this case. > >I think a way to encode that set in a very compact way will be to create a >perfect hash for just those positions and publish the function. (20 bits is all >that are needed and 800K entries at that size for a total of about 3 MB). > >Another possibility would be to have two bit vectors (one for losses and one for >draws). > >The nice thing about the bit vectors in a really sparse set is that they would >RLL compress a huge amount. Even if a 64 bit hash value were used, it would still be only 6,356,432 bytes Seems like a good idea for these real sparse ones.
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.