Information Technology Reference
In-Depth Information
theory has already been represented as rules and facts in Prolog). Therefore,
unification algorithm should be used to create an explanation structure. Based on
this, unification algorithm can be treated as the fundamental of realization of
EBG and the generalization can be formed.
The domain theory is stored as the inner database when EBG is implemented
based on Turbo Prolog. In terms of creating explanation structure of training
examples, it is easier to retrieval the database (domain theory) and search for the
matched rules like the way we use predicates.
Even if Prolog is based on unification algorithm, unification algorithm still
needs to be explicitly executed when implementing EBG with Prolog. This
means predicates of Turbo Prolog are required to finish the unification algorithm
and this is actually primitive idea of program design. Predicates are defined to
deal with term unification and term table unification. It is by means of unification
algorithm the explanation is accomplished and the proof tree is set up.
The next step of generalization is to get proof tree according to goal concept
regression. This regression involves substitution of constant with variables and
new term unification and so on. It is hard to effectively substitute constant with
variables, especially when the variable is not bounded. Though regression by
predicate has been defined, more work still need be done as to the way of
increasing its universality and treat with more variables not bounded.
Explanation and generalization are two steps of EBG. However it is easy to
think of explanation first and then the proof structure can be transferred to
generalization. Therefore, the two steps are executed consecutively and
independently, the whole proof tree and every path need to be preserved. Here,
we consider the two steps execute interchangeably. When the system attempts to
proof that the training example is an example of goal concept, search begins from
the goal concept backwardly until the training example gets a match. During this
process, every time a rule is used it is used to the variable goal concept without
example, thereafter the generalized explanation structure is constructed while at
the same time the explanation of training instances is created, the two phases of
EBG is combined together.
9.6.2
Meta Explanation
The simple Prolog meta explanation is as:
prolog(Leaf):-clause(Leaf,true).
prolog((Goal1,Goal2)):-
prolog(Goal1),
prolog(Goal2).
prolog(Goal):-
