Author: martin fierz
Date: 12:09:08 03/27/02
Go up one level in this thread
On March 27, 2002 at 13:21:48, Alvaro Jose Povoa Cardoso wrote: >His Martin, >Sorry I can't give any help on that matter. >I would apreciate some help though. >As you might know I also made a checkers engine (Portuguese/Spanish) and I >intend to do some EGTBs. >Could you please give me some insights on this? hi alvaro, i intend to make the source code of my generator public as soon as it's finished, and after i have added some comments. you should be able to adapt it easily then. but to answer your questions: >Specifically: > 1. What do you store on each EGTB entry? in uncompressed form, you store 4 values in a byte, win, loss, draw and unknown. > 2. Is a byte enough for an entry? which means that even in uncompressed form, you get 4 positions/byte > 3. How do we contruct an index from a position? this depends on how tricky you want to get. you can construct a 1-1 correspondance between index and position, which is what schaeffer did, or you can relax, and make a position->index which maps the N positions of a database on the numbers 1...N+M. of course, M should not be big, since you don't want to waste too much space. this is for example what ralph gasser did when solving 9 men's morris. i also did it this way - you can write a faster index function if you don't have to get an exact correspondence. > 4. Does a N piece TB refers to just N piece positions or could it refers to >less than N positions? (Or does it progresses to another TB of less that N >pieces?) ehm, i don't quite understand the question - and if i do, it is just a matter of terminology. if i say, i computed the 8-piece db, i mean i computed the 2,3,4,5,6,7, and 8piece db. > 5. Could you outline in some detail how does a TB generator works? i could, but there is actually a paper on this on schaeffers webpage, which explains everything in great detail: http://www.cs.ualberta.ca/~jonathan/Papers/Papers/databases.ps > 6. Dow you first generate all wins in 1 ply then in 2 the in 3...? something like that, yes. see that paper for details. > 7. Could you give me some examples of EGTBs file names? whatever you want? i'm using a naming scheme bm bk wm wk bmrank wmrank color, so for example 1322_34b.dat is the database which contains all positions with 1 black man, 3 black kings, 2 white men, 2 white kings, and the black man is confined to rows 1..3, the white men are confined to rows 1..4 (seen from the white side). and it's black to move. but in the end, you probably want to just use single files. > 8. How much EGTB files do you have at present? 2-7, and about 70% of the 8-piece db in positions, probably about 80% in terms of time - i expect it to finish in a week or so. it took about 2 days to compute the 2-7-piece db on an XP1600+ with 640MB ram, the 8-piece db is running since about 5 weeks. > 9. For a specifc EGTB, do we need to do a file for White and another for >black? Or can we use a single file for white and black? OR can we avoid this by >exchanging colors before lookup the TB? again, it's partly a matter of choice. i use separate files for white and black but there is nothing in principle that stops you from putting the 2 files together. if the database is symmetric, e.g. 3K-3K, you can forget about the side to move because you can change colors. aloha martin
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.