Author: GuyHaworth
Date: 00:43:41 07/13/05
Go up one level in this thread
I believe Eugene Nalimov evolved his code and index-scheme from Edwards': the latter can now be considered superceded. 'Bitbases' give Win/Draw/Loss data, and conceivably Win/Not_Win data only. I'm not up to date on generation time and compressed sizes but would have thought these quite good as a prelude to EGTs with values in. Vincent D is a fan, for one. The problem is that one might not secure the win, even if one is retaining it via bitbases. Nalimov's focus is to make the index as small as possible. He avoids 'unblockable checks' on the King not-to-move. This makes the index complex and 'unmoves' are difficult, which tends to slow an algorithm which is fully retrograde [uses 'unmoves'] like Thompson's, Wirth's or Wu's. Another issues around EN's EGTs is that the 'validation code' has about an 80% overlap with the 'generation code' which rather reduces its 'independence' which would be ideal in an 'independent verification' of the EGTs integrity. That said, I don't think EN has issued flawed EGTs on Rob Hyatt's site. I think there's another issue about the size of EN EGT indexes, but don't believe this is intrinsic to the nature of the index. The index could perhaps be retrieved as required from the EGT itself. An idea, if RAM allows, is to generate EGTs with a simpler index setup and then convert to Nalimov indexing later. The 'Chessmaster' FEG EGTs from J de Koenig seem to get generated faster than EN's with a different index arrangement which - however - makes validation next to impossible for 6-man EGTs at the moment. This did lead temporarily to an error lying undiscovered in these EGTs. They are only usable by Chessmaster which is another drawback. There are new ideas around which are generating 7-man EGTs with impressive speed. Runtime-compression and parallelism has not been fully investigated yet. The key difficulty is now one of storing and promulgating the EGTs after they have been created. g
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.