Author: Ricardo Gibert
Date: 20:36:25 09/09/99
Go up one level in this thread
On September 09, 1999 at 22:57:34, Andrew Slough wrote: >On September 09, 1999 at 21:57:28, Ricardo Gibert wrote: > >>Would it be feasible to generate an EGTB in the following manner: Instead of >>storing DTM or DTC in the EGTB, how about storing "nth generated move progresses >>towards mate"? It would seem this would be more compact in some endings. An old >>idea? Yes? No? Maybe? Incorporating a little primitive AI to order the moves >>generated would make the the tables even more compact. > >Information on the number of moves until a mate is needed during search probes >(there may have been 10 reversible moves, so I would want to know if I could >mate in 40) and during tablebase generation for algorithm termination. I'm not >convinced it would be a more compressed representation either, but then again >I'm not convinced Eugene's representation is optimal either... > >Move ordering is certainly the problem to make things compress well, but have >you ever watch a computer play KBBKN? I'm not sure that primitive AI would give >good results. > >Andy I figured out the problem with it. It would only be useful fot the more trivial EGs. With DTM for example, to generate the mate-in-4s, you only need the mate-in-threes, etc. so while overall it might be more compact, using it and generating it would be impractical. I have another half baked idea: Instead of DTM, use Distance-To-Mate-Modulo-N. So for example, in a KQKR ending with N=20 (or whatever it actually is), the EGTB would return mate in 7 or 27. For sufficiently large N this would not result in ambiguity, since a sub-optimal move would either throw away the win completely or delay the win by at most N-1 moves. This would reduce the size of the EGTB. The proper choice for N would need to be determined from DTM EGTB tables. Generating the DTMMN table would require a trivial folding of DTM. I would not be surprised if DTM already does this.
This page took 0.01 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.