Author: GuyHaworth
Date: 00:43:53 11/23/05
Go up one level in this thread
Certainly, the function of accessing whatever portfolio of EGTs were available is separate from: - the core search-engine's function, and - the function of accessing an individual EGT-type and EGT Therefore, it should be functionally separate in the software architecture. One could therefore imagine this module evolving behind defined interfaces, but quite independently of the FRITZ, FRUIT, JUNIOR, SHREDDER etc developments. Such a module might have access to EGTs of varied format and metric: - deKoenig (FEG), Nalimov (most promulgated), Konoval (to come) - DTC(onversion), DTM(ate) (Nalimov), DTZ (most compact), WDL (bit-pair map) Further, such a module might be more dynamic at runtime. EGTs are consuming many GBytes now, and there is an argument now for saving space and using time as and when required instead ... EGTs can be 'cut back' to hold, say, only the draws, broken positions and wins/losses 0 and 1 moves from the game/subgame boundary. It could then be reconstructed more rapidly because the draws are known and there are no required references to subgames. Reconstructing the EGT would be an integral part of 'forward search', done differently when it was known that definitive data could be 'backed up' to evaluate the position. The speed demonstrated by Yakov Konoval's program, not yet parallelised, shows that this could be done at runtime within a game for 5-man positions and some 6-man. The potential use of FREEZER within-game is also possible, and maybe SHREDDER is there already. 6-man positions with a pair of blocked-pawns are effectively 4-man EGT-generations with 5-man EGT lookup. How long it would take would depend on the depth of the position to be analysed in the game. Although DTx-graphs often have long tails, average depths are usually much less than maxDTx. Incidentally, this would raise an interesting philosophical question for man-machine matches where the carbon side objects to the silicon having access to N-man EGTs. "No, we have no N-man EGTs - but we do have the ability to create them as a legitimate part of the search function." 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.