Next: Other Brute Features
Up: Learning Sets of Rules
Previous: Design Choices in Learning
- greedy search can miss good rules
- disallowing overlapping rules can cause problems
- so....look at every rule and if it is good keep it
- uses depth bound
- overfitting is *more* of a problem
- when evaluating single rules coverage is important
- we used chi-square statistic as our performance measure
- validation sets are difficult - we used randomization testing
- still is an issue with similar rules and additional conjuncts
- presenting multiple rules are difficult
Patricia Jean Riddle
Wed Jun 23 13:06:34 NZST 1999