ifix

packages/ProbLog/CMakeLists.txt

@@ -5,6 +5,7 @@ set (PROGRAMS
   dtproblog.yap 
   aproblog.yap 
   problog_learning.yap
+  problog_lbfgs.yap
   problog_learning_lbdd.yap
   )
 

packages/ProbLog/problog_lbfgs.yap

@@ -1,4 +1,4 @@
-%xb%%% -*- Mode: Prolog; -*-
+%%% -*- Mode: Prolog; -*-
 
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 %
@@ -285,6 +285,7 @@ save_model:-
 %=
 %========================================================================
 
+
 check_examples :-
 	%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 	% Check example IDs
@@ -602,7 +603,8 @@ init_one_query(QueryID,Query,Type) :-
 	->
 	 format_learning(3,' Reuse existing BDD ~q~n~n',[QueryID])
 	 ;
-	  b_setval(problog_required_keep_ground_ids,false),
+	 b_setval(problog_required_keep_ground_ids,false),
+	 (QueryID mod 100 =:= 0 -> writeln(QueryID) ; true),
 	  problog_flag(init_method,(Query,N,Bdd,graph2bdd(X,Y,N,Bdd))),
 	  Query =.. [_,X,Y]
 	  ->
@@ -915,13 +917,10 @@ bind_maplist([Node-Theta|MapList], Slope, X) :-
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 % start calculate gradient
 	%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-user:evaluate(L, X0,Grad,N,_,_) :-
+user:evaluate(LLH_Training_Queries, X,Grad,N,_,_) :-
     Handle = user_error,
     problog_flag(sigmoid_slope,Slope),
-		forall(between(0,N1,I),
-	   (V is random, X[I] <== V) %, sigmoid(X[I],Slope,Probs[I]) )
-	   )
-    Probs = X, %<== array[N] of floats,
+    Probs = X,
     N1 is N-1,
     forall(between(0,N1,I),
 	   (Grad[I] <== 0.0) %, sigmoid(X[I],Slope,Probs[I]) )
@@ -931,12 +930,10 @@ user:evaluate(L, X0,Grad,N,_,_) :-
 	    LLs
 	   ),
     sum_list(LLs,LLH_Training_Queries),
-	forall(tunable_fact(FactID,GroundTruth), (Z<==X[FactID],W<==Grad[FactID])),
-    L = LLH_Training_Queries.
+	forall(tunable_fact(FactID,GroundTruth), (Z<==X[FactID],W<==Grad[FactID],writeln(FactID:(W->Z)))).
 
 compute_grad(N, X, Grad, Probs, Slope, Handle, LL) :-
     user:example(QueryID,Query,QueryProb,Type),
-    recorded(QueryID, BDD, _),
     BDD = bdd(Dir, GradTree, MapList),
     bind_maplist(MapList, Slope, Probs),
     qprobability(BDD,Slope,BDDProb),
@@ -947,8 +944,8 @@ gradientpair(BDD,Slope,BDDProb, QueryProb, Grad) :-
     qgradient(BDD, Slope, FactID, GradValue),
     % writeln(FactID),
     G0 <== Grad[FactID],
-    GN is G0-GradValue*(QueryProb-BDDProb),
-    %writeln(FactID:(G0->GN)),
+writeln(    GN is G0-GradValue*(QueryProb-BDDProb)),    GN is G0-GradValue*(QueryProb-BDDProb),
+    writeln(FactID:(G0->GN)),
     Grad[FactID] <== GN.
 gradientpair(_BDD,_Slope,_BDDProb, _Grad).
 

packages/ProbLog/problog_learning_lbdd.yap

@@ -624,7 +624,6 @@ init_one_query(QueryID,Query,Type) :-
         %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 	% if BDD file does not exist, call ProbLog
 	%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-	  writeln(QueryID),
 	(
 	 recorded(QueryID, _, _)
 	->
@@ -659,7 +658,6 @@ init_one_query(QueryID,Query,Type) :-
 	  maplist_to_hash(MapList, H0, Hash),
 	  Tree \= [],
 	  %put_code(0'.),
-	  writeln(QueryID),
 	  tree_to_grad(Tree, Hash, [], Grad),
 	 recordz(QueryID,bdd(Dir, Grad, MapList),_)
 	).
@@ -739,7 +737,6 @@ gradient(QueryID, l, Slope) :-
 	bind_maplist(MapList),
 	run_sp(Tree, Slope, 1.0, Prob0),
 	(Dir == 1 -> Prob0 = Prob ;  Prob is 1.0-Prob0),
-%writeln(QueryID:Prob),
 	assert(query_probability_intern(QueryID,Prob)),
 	fail.
 gradient(_QueryID, l, _).

packages/yap-lbfgs/lbfgs.pl

@@ -156,10 +156,8 @@ yes
 
 
 /** @pred lbfgs_initialize(+N, -SolverInfo)
-The same as before, except that the user module is the default
-value.
 
-Example
+Do initial memory allocation and a reference to a descriptor.
 ~~~~
 lbfgs_initialize(1, Block)
 ~~~~~
@@ -169,26 +167,25 @@ lbfgs_initialize(N,X,t(N,X,U,Params)) :-
 
 lbfgs_initialize(N,X,U,t(N,X,U,Params)) :-
     lbfgs_defaults(Params),
-	integer(N),
-	N>0,
-
-	% check whether there are such call back functions
-
-	lbfgs_grab(N,X).
+    integer(N),
+    N>0,
+    lbfgs_grab(N,X).
 
 	% install call back predicates in the user module which call
 	% the predicates given by the arguments
 
 
-/** @pred  lbfgs_finalize/0
+/** @pred  lbfgs_finalize(+State)
+
 Clean up the memory.
 */
 lbfgs_finalize(t(N,X,U,Params)) :-
 	lbfgs_release(X) ,
 	lbfgs_release_parameters(Params) .
 
-/** @pred  lbfgs_run/2
-Do the work.
+/** @pred  lbfgs_run(+State, -FinalOutput)
+
+run the algorithm. output the final score of the function being optimised
 */
 lbfgs_run(t(N,X,U,Params),FX) :-
     lbfgs(N,X, Params, U, FX).