Author: Mark Longridge
Date: 12:26:30 01/27/01
We do something very similar to a tb in Rubik's Cube searches, we search for optimal solutions to a specific position. We can't search the whole tree (4.3x10^19 positions), but it's still interesting to find the optimal moves. Why not make a partial tb for some interesting positions, where the starting position is fixed, like we do for rubik's cube, and calculate that to the end. Eventually you can go back and add more positions later. I'm thinking this could work well with pawn only positions, like kppkppp or something similar. I don't think calculating this to the end would be too hard, or at least calculating to a winning position. Eugene, what do you think? Could tbgen be easily modified to do this, or is it not practical? Surely starting from one node with a pawns only position shouldn't be too bad. Instead of searching back from mate, you could search back from a canidate winning position. My second idea was to hike up the position learning from 64K to something a lot bigger for crafty. It doesn't seem to take too long to search through position.bin, ditto for book learning... make it so it can learn new lines. There doesn't seem to be any real limit on how big book.bin can get, but it seems to me that my book.bin isn't getting any bigger, so I don't see where it puts information on completely new opening moves. A mate.bin might be an idea too. All the really deep mates which occur in real games could be put in mate.bin, 24 hours of computer analysis could figure out a lot deep mates from fixed positions. Mark
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.