Artificial Intelligence: Expert Systems with CLIPS and Free Software Maximizing Your Fishing Profitability with a rod and reel off the waters of NYC

December 2015

Artificial Intelligence: Expert Systems with CLIPS and Free Software

LIU/Brooklyn Artificial Intelligence

Fishing Season Schedule strategy

In evaluating the schedules for fishing seasons it becomes apparent very quickly how unwieldy this data tree and the resulting code base will become, even with just a moderate number of species, and only two jurisdictions. If I add neighboring states, and the mostly ignored reciprocity laws (what happens when a New York boat fishes in New Jersey waters) it can take a team of very smart people to create the correct decision tree. Fortunately, with the use of the inference engine in CLIPS, we can do things in a slightly different way.

There are two ways to approach this decision tree. First and most obvious is to handle it like a series of nodes, each one triggering the next according to response. CLIPS is an inference engine and such a solution is suggested by Giarratano & Riley in there text on CLIPS, "Expert Systems, Principles and Programming". In this case, rules are created where the user is prompted to give an answer questions which cause the CLIPS agenda to start the next appropriate rule until a leaf is reaches with an answer. A quick mock up of this kind of code is as follows:

(deftemplate node
	(slot name )
	(slot type )
	(slot question )
	(slot yes-node)
	(slot no-node )
	(slot answer )
)

This is similar to what we would see as a structure in procedural code.

(deffacts tree
	(node (name root)(type decision)
		(question "Is the animal warm blooded?"  )
		(yes-node node1) (no-node node2))
	(node (name node1)(type decision)
		(question "Does it purr?")
		(yes-node node3) (no-node node4))
	(node (name node2) (type answer) (answer snake))
	(node (name node3) (type answer) (answer cat))
	(node (name node4) (type answer) (answer dog))
	(current-node root)
)

These create nodes on reset that either push you down the tree (decisions) or are leafs (answer). Now we just need some rules to stitch it together.

(defrule ask-decision-node-question
	?node <- (current-node ?name)
	(node (name ?name)
			(type decision)
			(question ?question))
			(not (answer ?))
=>
	(printout t ?question "(yes or no) ")
	(assert (answer (read)))
)
;*******
;******
;*****
(defrule bad-answer
	?answer <- (answer ~yes&~no)
=> 
	(retract ?answer)
)

(defrule proceed-to-yes-branch
	?node <- (current-node ?name)
	(node (name ?name)
			(type decision)
			(yes-node ?yes-branch))
   ?answer <- (answer yes)
  =>
	(retract ?node ?answer)
	(assert (current-node ?yes-branch))
)
;*******
;******
;*****
				  
(defrule proceed-to-no-branch
	?node <- (current-node ?name)
	(node (name ?name)
			(type decision)
			(no-node ?yes-branch))
   ?answer <- (ansert no)
  =>
	(retract ?node ?answer)
	(assert (current-node ?no-branch))
)

One set of rules for yes decisions, another for no decisions and a third
bad answers.  The last rule is for the leaf

(defrule correct
	?node <- (current-node ?name)
	(node (name ?name) (tyoe answer)(answer ?value))
	(not (answer ?))
					  =>
	(prinout t "I guess it is a " ?value crlf)
)

At this point I note three things here:

This code is an adaptation of the code example in the Giarratano & Riley text noted above.
I have not tested this code and leave that to you.
The complete code in the text is far boarder than this and worth looking into and testing in its own right. I encourage you to get a copy of the text.

As we examine this, we can see that this will allow us to create our tree exactly like our diagram, but in this case, it is a waste of a good inference engine. It assumes a single species result. We have results of several species at once and then each of those branches need to be tested for all possible weather conditions, which is a lot of repeating code, and code that repeats itself is a clear indication that the approach to the problem is not correct.
So we have a better approach