3.5.2 2b: Dumb Wumps agent

This week, we’ll add an automatic wumpus walker. It will be pretty stupid. We’ll make it smarter next week.

Work in a new directory (2/b) and start by copying your 2/a code into 2/b.

To do:

• Get rid of the all the globals in GTF. Replace them with a working memory (hereafter, wme) consisting of...

  (defstruct wme edges nodes pos visited)
  

  Now add a w var that is initialized at start of game and carried around all the wumpus code.

• Convert GTW from manual to automatic. So the code now longer reads input from the REPL by a function that asks "where can i go?" and selects one direction, as per the next point.
• Dumb wumpus gets a set of nodes and orders them by some value heuristic. Then selects the top one of that order (if ties, then it picks at random). Given the rules of the game, given a set of out edges from one node, implement a sort order on those edges.

Internally, this will be some multi-utility functions that does feature extraction of each edge (sirens? glow? stink?, whatever, use your creativity) and weights these is some way; e.g.

; something like this. note: written, not tested
(defun what2DoNext (node)
    (let* ((edges  (edges-from-this-node node))
           (scored (mapcar #'edgeScore edges))
           (sorted (sort scored #'> :key #'car)))
      (second (first sorted))))

(defun edgeScore (e)
    (let ((sirens? (hasSirens  e))
          (glow?   (hasGlow    e))
          (etc?    (otherStuff e)))
       (cons (edgeScoreFn sires? glow? etc?) e)))
    
(defun edgeScoreFn (sirens? glow? etc?)
    "a b c  are weights on different heuristic search measures"
     (+ (* a (if sirens? 1 0) (zip))
        (* b (if glow?   1 0) (zip))
        (* c etc)             (zip)))

(defun zip ()
    "Returns a random number between 0.99 and 0.999.
     Enables random selection."
    (+ 0.99 (randf 0.009))
)

What to Hand In

As before, report any changed code.

Also, hand in a REPL session of dumb wumpus playing the game. Include enough trace info so the REPL session reveals what is going on. Include displays of moves, deaths, captures, etc in the REPL text.

Text files (short) describing your edge ranking heuristics.

This document was generated on March 1, 2011 using texi2html 5.0.