Algorithm 1.

Covering Strategy

Input: E+ (set of positive examples), E− (set of negative examples)

Output: H (A set of GEP-based rules)

1:  /* Initialization */

2:  H ← ∅

3:  Lmin ← +∞ (minimum description length obtained)

4:  LH ← 0 (current description length)

5:  Ltheory ← 0 (theory bits)

6:

7:  /* Learning */

8:  Repeat

9:          Learn a rule R to cover the positive samples in E+

10:        E+ ← E+−{s ∣ s can be covered by R}

11:        /* Pruning */

12:        Ltheory ← Ltheory + number of bits for encoding R

13:        Lexception(H) ←number of bits for encoding current exceptions

14:        LH ← 0.5·Ltheory + Lexception(H)

15:        If (LH < Lmin) Then

16:            H ← H∪{R}

17:        Else

18:            Termination

19:        /* Update */

20:        If (Lmin > LH) Then

21:            Lmin ← LH

22:  Until E+ = = ∅