Reduced Cost Ordering: Postprocessor for FOCL
Input
- n sets of clauses (one per class)
- classified examples (distinct from examples used to create clauses)
- Cost matrix
Output
- Ordered decision list using subset of the clauses
- default rule
1. Let Pool = an unordered set of rules learned on training set
2. Initialize Rules to a default (Least Expected Cost Class)
3. Add single rule (and new default) to Rules that most
lowers misclassification cost on ordering set.
4. Repeat until addition wouldn’t lower misclassification cost.