Decision Lists and ILP
FOCL (and FOIL, etc.) designed for binary classification problems
- learn clauses that cover all positive an no negatives
When there are n classes, run learner n times
- treat examples of each class in turn as positives
- treat examples of other classes as negative
Classification issues
- When clauses from more than one class are satisfied
- When no clause is satisfied
Decision Lists
- Ordered set of clauses
- Default clause if no clause matches
Better Safe Than Sorry (Provost & Buchanan)