Author: Vincent Diepeveen
Date: 18:39:06 02/02/01
Go up one level in this thread
On February 02, 2001 at 14:43:39, J. Wesley Cleveland wrote: >On February 01, 2001 at 21:02:42, Vincent Diepeveen wrote: > >>On January 31, 2001 at 19:14:19, Dann Corbit wrote: >> >>>On January 31, 2001 at 11:36:01, Robert Hyatt wrote: >>>[snip] >>>>The idea seems pretty good for 3-4 piece files, and even for 5 piece files >>>>although the memory to hold them becomes prodigous. But 6's are hopeless >>>>as todays machines are no better at probing a 1gb file than they are at >>>>probing a 1 terrabyte file. >>> >>>It might be worthwhile to store them in a real database with hashed index. >>> >>>Modern database systems will cache database requests very efficiently, and so I >>>think it might be doable. The database model would be the hard part (finding a >>>representation which is still highly compact like the tablebase files) >> >>If you don't store it in 2 bits but only win/draw/loss (in diep >>i only generate legal moves so i can never get to illegal positions) >>then you can store 5 positions instead of 4 in a single byte. >> >>Those compress very good. experiments of mine so far come down to >>the size of 1 CD of 650mb for all 3,4,5 men. >> > >I would guess that after compression the 4 per byte tables would be only >slightly larger than 5 per byte tables, and might not be worth the extra >decoding effort. about 20%-25% more space you need for 4 per byte as for 5 per byte if i may guess. So for a very simplistic step you save 20% that's worth it of course! And it doesn't eat a 15mb RAM table or something which takes 30 seconds to get generated at program startup! So it sure is worth doing it! >>Your first big safe is the size in which you store. the second big big >>safe is the better compression. >> >>Of course you can get them down even quite smaller but then you can't >>quickly read them during a game unless you keep everything in RAM as >>you need to do binary search on the positions then, >>which no user except a few freaks with the latest >>computers will do. >> >>The problem remains however generation. Also you'll not win KNNKP >>that's the only price you pay when talking about 5 men :) > >If you have the HD space, it would be better to have both WLD and full tables, >and always use the full table at the root. HD space isn't the problem at all for generation. To generate a 6 men like the nightmare ones with just 1 pawn and not the same pieces of same side one needs only a few tens of GIGABYTES in total for generation at most as after generation they all get compressed so bigtime that even the conversion to other 6 men only takes very little compared to the uncompressed bitmaps and other stuff for generation. However the generation is the real problem, apart from needing near to a gigabyte of RAM at least... ...at this moment i don't have a computer left to generate them, not to mention 80 or 40mb/s SCSI harddisks Generation of a single 6 men with 1 pawn is a nightmare, it is taking between 1 and 2 weeks assuming a gigabyte of RAM... ...and a very fast harddisk and cpu... Greetings, Vincent
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.