gros1530@fredonia.UUCP (Dan Gross) (03/21/90)
In the AI course I'm currently enrolled in, we recently discussed
Game Theory. When we hit Knuth's and Moore's NEGMAX procedure, my other-
wise scholarly professor balked. Unfortunately the text we are using
is rather cryptic on this topic, so I bring my question here...
Is the NEGMAX procedure meant to to work with 1,0,-1 evaluation
ONLY, or can it work with any evaluation with -# for loss and +# for win?
In other words, will it work if the terminal nodes are something besides the
basic -1,0,1 (ex. -2.5 -.97 .98 .34 -1.3)?
Just curious,
--Dan Gross
--
Dan Gross fredonia!gros1530@cs.buffalo.edu
430 Washington Ave. UUCP:...{ucbvax,rutgers}!sunybcs!fredonia!gros1530
Dunkirk, NY 14048-2121 Curiosity may have killed the cat,
716-366-0405 but at least he didn't die ignorant!jgk@osc.COM (Joe Keane) (03/22/90)
In article <1778@fredonia.UUCP> fredonia!gros1530@cs.buffalo.edu (Dan Gross) writes: > Is the NEGMAX procedure meant to to work with 1,0,-1 evaluation >ONLY, or can it work with any evaluation with -# for loss and +# for win? >In other words, will it work if the terminal nodes are something besides the >basic -1,0,1 (ex. -2.5 -.97 .98 .34 -1.3)? Yes, it works fine with continuous evaluations. The difference between it and MiniMax is minimal; you just flip the signs on every other ply. It just comes down to which is easier to code, or runs faster.