Author: Chris Carson
Date: 05:49:11 07/29/99
Ok, here is a thought (perhaps not a good one). ;) Background: I will refer to Crafty as an example. Crafty has an "Opening" book that is based on actual PGN games with factors that help the program navigate the opening. Crafty has "Positional" learning that will store evaluations of a position for later reference. This is short term learning. I am thinking of long term or permanent knowledge. Most programs are very good at tactics, but "Positional" evaluation is considered by many to be lacking in most programs. Suggestion: Can a position (FEN) be analyzed by a human (like LCTII tests) and a good "Positional" or several (rank order) "Positional" moves be be identified? It would look like the "Positional" learning in Crafty, but the program would never modify/delete the references (this would be a permanent "Positional" book). The program would check the "Positional" book (just like it does for the "Opening" book) and if a "Position" was present, then it would select the recommended move (just like crafty "Positionl" learning) with the know evaluation. This is different from "Positional" learning since a person did the evaluation and the program will not create/modify/delete the book during a game (like the "Opening" book). The idea is to add a "Positional" book that would provide the program with "Postional" knowledge. Perhaps "Positional" test suites with known "Best Moves" (bm) would be a good start (the "Positions" from that the programs can not easily < 3min solve). The "Positional" book may grow to be very large. Thus I would recommend that only hard to solve "Positional" cases be in the book and not cases with tactical solutions. Thoughts? Best Regards, Chris Carson
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.