Author: Eugene Nalimov
Date: 22:13:09 09/04/99
Go up one level in this thread
Probably you should look at TBINDEX.CPP, and try to understand indexing code. I know that's hard (several people gave up :-)), but you can at least try. Hopefully I and Ernst Heinz will write the paper about TBs indexing. You can look into ICCAJ - this year there was an excellent Ernst's article about prior state of the art. I was able to improve it by (1) removing some invalid positions (king of the side not to move is checked, and checking piece is on the square that is immediately near king's square, so it cannot be blocked), (2) noticing the fact that when you put the pawn on the board, usually there are less than 48 possible squares, as one or both kings are on A2-G7, and (3) implementing good EP schema. Possible improvement that I understood too late to implement in TBGEN is related to (2) - it's possible to optimize some of the cases I don't optimize now. Maybe for 6 man TBs... Eugene On September 04, 1999 at 04:32:48, Michel Langeveld wrote: >On September 03, 1999 at 15:07:09, Eugene Nalimov wrote: > >>Using indixing schema that is currently used by TBGEN: >> >>KQQKQQ: 1,237,357,440 for each side to move. In this case, we need only one >>file, as black material is identical to white material. So, 1,237,357,440 >>positions total. >> >>KQQKQP: wtm, 7,739,363,232 positions; btm, 8,024,974,680 positions (that is >>assuming you will not implement additional enumeration tables - that's possible, >>but will eat a lot of RAM). Total 15,764,337,912 positions. >> >>Eugene > >Thanks for the answer I shall try to recalculate to the same number. >Maybe if I understand fully the theory I can help with some addition ideas. >I understand for KQQKQQ only one file is needed. I checked all my 4 piece >tablebases which is ... bytes total. > >The amount of KQKQ.nbb, KRKR.nbb, KBKB.nbb, KPKP.nbb = 10.341.448. So these >files are not necesarry when the tablebasehit algorithm is adapted a little. >Amazing that this is 10.3/146.0 = 7% smaller. > >I remember that you posted once in the Crafty mailinglist that you could drop >the total amount of bytes of the current set of .nbw and .nbb a slightly. Did >you refer to this above or something else? If something else then can you >explain it? > >Michel Langeveld
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.