RDM Chapter 6: Upgrading

1
RDM Chapter 6 Upgrading

• prepared for COMP422/522-2008, Bernhard Pfahringer

2
Upgrading propositional algos

• Extend conservatively to keep propositional as
  special case (examples Foil, Tilde, WarmR)
• change example representation
• change hypothesis representation
• modify search space traversal (biases)
• add features, e.g. background knowledge

3
Foil

• see copies from first lecture
• standard separate-and-cover rule learner
• (problem with greediness add determinate
  literals)

4
Tilde

• straightforward decision tree learner
• slight complication with variables
• see Fig.6.3. and Algorithm 6.3

5
WarmR

• find key(X) ? r1(X,),r2(),
• frequent patterns
• freq(c,D) ? D ? c key?
• relational association rule
• a IF q THEN l
• when support(a)s can confidence s and (key ?
  q) and (key ? q,l)
• see Algorithm 6.4 (level-wise, Apriori)
• important optimization 6.8 (but not for
  relational yet)

6
Language Bias

• needed for efficiency, tractability
• syntactic types and modes (example 6.15), full
  grammar (ex 6.16)
• semantic (in)determinate clauses (ex. 6.19),
  functional clauses (ex 6.20)