Conflict Resolution

Take last week's semantic nets labs and copy it to a new file.
Remove the facts in the semantic net file so that you have the smallest number of facts that you need to override inheritence.
If you load and run the system, it should still give you the correct results, however if the rules were applied in a different order, they would not override the inheritence.
In what order are the rules applied? You can use the step menu item, Ctl-T, or (run 1) to step through.
You will note that many rules are applicable at most stages. (Use (agenda) to see which rules). The strategy that is used can be modified.
You can see which strategy is being used by (get-strategy), and change the strategy by (set-strategy new-strat)
I've found that the default strategy is depth. How does the order of rule application change if you use breadth? Does the result change?
Unfortunately, I've not found a simple mechanism that Clips gives us for backward chaining.
That's too bad, because we don't always want to store all the information about a particular object, but would like to derive it at runtime.
For instance, my net has the feature of how the animal travels. I can derive all the facts (cache them as it were), or I can keep the basic net structure and derive the fact when I need it.
Deriving as few new facts as possible write some rules that enable you tell if x ISA y. I use an assertion (IS? bat mammal) to drive this.
I took advantage of the salience property of a rule which allows you to rank which rules apply first. (defrule A (declare (salience 5)) LHS => RHS). In essence this overrides the conflict resolution strategy.
Make sure that you also can say if x ISnotA y.
Deriving as few new facts as possible write some rules that enable you tell if what value x has for feature y. I use an assertion (feature? bat blood) to drive this.
Now load in the full semantic net from lab 5, and see how the system performs on a range of questions.