ontology_induction_generated.pl

:- dynamic member/2, induced_rule/2, cooccur/3.

% --- Background co-occurrence facts (generated by XSLT from ontology.xml) ---
cooccur('stack','stack_op',1).
cooccur('dup','stack_op',3).
cooccur('drop','stack_op',3).
cooccur('swap','stack_op',3).
cooccur('over','stack_op',2).
cooccur('operations','stack_op',2).
cooccur('dictionary','dict_op',2).
cooccur('words','dict_op',3).
cooccur('contains','dict_op',3).
cooccur('holds','dict_op',0).
cooccur('compiler','compiler_word',2).
cooccur('compiles','compiler_word',2).
cooccur('colon','compiler_word',2).
cooccur('semicolon','compiler_word',3).
cooccur('defined','compiler_word',1).
cooccur('interpreter','interpreter_word',1).
cooccur('interprets','interpreter_word',1).
cooccur('pocketforth','engine_word',0).
cooccur('engine','engine_word',2).
cooccur('pocket','engine_word',1).
cooccur('executes','engine_word',2).
cooccur('inside','engine_word',1).
cooccur('harness','engine_word',0).
cooccur('transformer','meta_word',0).
cooccur('xslt','meta_word',2).
cooccur('asp','meta_word',3).
cooccur('xml','meta_word',3).
cooccur('code','meta_word',3).
cooccur('generates','meta_word',2).
cooccur('validates','meta_word',3).
cooccur('architecture','meta_word',3).
cooccur('constraints','meta_word',1).
cooccur('learns','learning_word',5).
cooccur('statistical','learning_word',3).
cooccur('relationships','learning_word',5).
cooccur('text','learning_word',1).
cooccur('cooccurrence','learning_word',2).
cooccur('pmi','learning_word',2).
cooccur('embeddings','learning_word',6).
cooccur('neural','learning_word',3).
cooccur('network','learning_word',4).
cooccur('capture','learning_word',2).
cooccur('similarity','learning_word',1).

% --- Concept hierarchy ---
concept(forth_word).
is_a(forth_word, root).
concept(stack_op).       is_a(stack_op,       forth_word).
concept(dict_op).        is_a(dict_op,        forth_word).
concept(compiler_word).  is_a(compiler_word,  forth_word).
concept(interpreter_word). is_a(interpreter_word, forth_word).
concept(engine_word).    is_a(engine_word,    forth_word).
concept(meta_word).      is_a(meta_word,      forth_word).
concept(learning_word).  is_a(learning_word,  forth_word).

% --- Seed members ---
member('stack',       stack_op).
member('dup',         stack_op).
member('drop',        stack_op).
member('swap',        stack_op).
member('over',        stack_op).
member('operations',  stack_op).
member('dictionary',  dict_op).
member('words',       dict_op).
member('contains',    dict_op).
member('compiler',    compiler_word).
member('compiles',    compiler_word).
member('colon',       compiler_word).
member('semicolon',   compiler_word).
member('defined',     compiler_word).
member('interpreter', interpreter_word).
member('interprets',  interpreter_word).
member('pocketforth', engine_word).
member('engine',      engine_word).
member('pocket',      engine_word).
member('executes',    engine_word).
member('xslt',        meta_word).
member('asp',         meta_word).
member('xml',         meta_word).
member('generates',   meta_word).
member('validates',   meta_word).
member('learns',      learning_word).
member('statistical', learning_word).
member('relationships', learning_word).
member('embeddings',  learning_word).
member('neural',      learning_word).
member('network',     learning_word).

% --- ILP Engine (generated by XSLT from ontology.xml) ---
candidate_score(Concept, Context, Precision, Recall, F1, PosCovered) :-
    findall(W, member(W, Concept), Pos),
    length(Pos, PosTotal),
    PosTotal > 0,
    findall(W, (member(W, Concept), cooccur(W, Context, _)), Covered),
    length(Covered, PosCovered),
    findall(W, cooccur(W, Context, _), AllWithContext),
    length(AllWithContext, AllTotal),
    (AllTotal > 0 -> Precision is PosCovered / AllTotal ; Precision = 0),
    Recall is PosCovered / PosTotal,
    (Precision + Recall > 0 -> F1 is 2*Precision*Recall/(Precision+Recall) ; F1 = 0).

best_rule_for_concept(Concept, BestContext, BestF1) :-
    setof(Context, W^Cnt^cooccur(W, Context, Cnt), Contexts),
    findall((F1,Context), candidate_score(Concept, Context, _, _, F1, _), Scores),
    Scores \= [],
    max_member((BestF1, BestContext), Scores).

induce_rules :-
    retractall(induced_rule(_,_)),
    forall(concept(C),
        ( best_rule_for_concept(C, Context, F1),
          F1 > 0.2,
          assertz(induced_rule(C, Context)),
          format("Induced: is_a(W, ~w) :- cooccur(W, '~w'). F1=~2f~n", [C, Context, F1])
        ; format("No rule for ~w (insufficient evidence)~n", [C])
        )).

propose_new_members :-
    forall(
        ( cooccur(W, Context, Cnt), Cnt >= 1,
          induced_rule(Concept, Context),
          \+ member(W, Concept),
          \+ W = Concept ),
        format("Propose: ~w should be is_a(~w) because cooccurs with '~w' cnt=~w~n",
               [W, Concept, Context, Cnt])).

generate_ontology_xml :-
    format("<ontology name=\"Induced-Next\">~n"),
    forall(concept(C),
        ( induced_rule(C, Ctx),
          format("  <concept id=\"~w\" is_a=\"forth_word\" induced_context=\"~w\">~n", [C, Ctx]),
          forall(member(W,C), format("    <member word=\"~w\"/>~n", [W])),
          format("  </concept>~n")
        )),
    format("</ontology>~n").

:- format("=== PocketLearn ILP Induction (XSLT-generated from PocketForthConcepts) ===~n"),
   induce_rules,
   format("~n=== Proposing new members ===~n"),
   propose_new_members.