Following is the pseudo code for PDL2's learning
algorithm. It is executed for each training pair q. The
relations equal (eq), subset (sub) and superset (super)
are the usual ones applied to the sets of examples
represented by pairs. The relation discriminated
(discr) means that the sets are disjoint. Two pairs that
differ in exactly one asserted input value are near
match (n_match), and num_assert is the number of
asserted inputs in a pair. Finally, two pairs are
concordant (conc) if they have the same output value.