This repository has been archived on 2023-08-20. You can view files and clone it, but cannot push or open issues or pull requests.
yap-6.3/packages/cplint/lemur/slipcover_lemur.pl
Fabrizio Riguzzi b25c9e5b61 lemur
2014-10-15 15:15:46 +02:00

1411 lines
33 KiB
Prolog

/*
SLIPCOVER
Copyright (c) 2013, Fabrizio Riguzzi and Elena Bellodi
*/
:-use_module(library(lists)).
:-use_module(library(random)).
:-use_module(library(system)).
:-dynamic setting/2,last_id/1, rule/5.
:- op(500,fx,#).
:- op(500,fx,'-#').
:-[revise_lemur].
setting(epsilon_em,0.0001).
setting(epsilon_em_fraction,0.00001).
setting(eps,0.0001).
setting(eps_f,0.00001).
/* if the difference in log likelihood in two successive em iteration is smaller
than epsilon_em, then EM stops */
setting(epsilon_sem,2).
/* number of random restarts of em */
setting(random_restarts_REFnumber,1).
setting(random_restarts_number,1).
setting(iterREF,-1).
setting(iter,-1).
setting(examples,atoms).
setting(group,1).
setting(d,1).
setting(verbosity,1).
setting(logzero,log(0.000001)).
setting(megaex_bottom,1).
setting(initial_clauses_per_megaex,1).
setting(max_iter,10).
setting(max_iter_structure,10000).
setting(maxdepth_var,2).
setting(beamsize,100).
setting(background_clauses,50).
%setting(specialization,bottom).
setting(specialization,mode).
/* allowed values: mode,bottom */
setting(seed,rand(10,1231,30322)).
setting(score,ll).
/* allowed values: ll aucpr */
sl(File):-
setting(seed,Seed),
setrand(Seed),
generate_file_names(File,FileKB,FileIn,FileBG,FileOut,FileL),
reconsult(FileL),
load_models(FileKB,DB),
assert(database(DB)),
statistics(walltime,[_,_]),
(file_exists(FileBG)->
set(compiling,on),
load(FileBG,_ThBG,RBG),
set(compiling,off),
generate_clauses(RBG,_RBG1,0,[],ThBG),
assert_all(ThBG)
;
true
),
(file_exists(FileIn)->
set(compiling,on),
load(FileIn,_Th1,R1),
set(compiling,off)
;
(setting(specialization,bottom)->
setting(megaex_bottom,MB),
deduct(MB,DB,[],InitialTheory),
length(InitialTheory,_LI),
remove_duplicates(InitialTheory,R1)
;
get_head_atoms(O),
generate_top_cl(O,R1)
)
),
% write('Initial theory'),nl,
% write_rules(R1,user_output),
learn_struct(DB,R1,R2,Score2),
learn_params(DB,R2,R,Score),
statistics(walltime,[_,WT]),
WTS is WT/1000,
format("~nRefinement score ~f - score after EMBLEM ~f~n",[Score2,Score]),
format("Total execution time ~f~n~n",[WTS]),
write_rules(R,user_output),
listing(setting/2),
open(FileOut,write,Stream),
format(Stream,'/* SLIPCOVER Final score ~f~n',[Score]),
format(Stream,'Execution time ~f~n',[WTS]),
tell(Stream),
listing(setting/2),
format(Stream,'*/~n~n',[]),
told,
open(FileOut,append,Stream1),
write_rules(R,Stream1),
close(Stream1).
gen_fixed([],[]).
gen_fixed([(H,B,BL)|T],[rule(R,H,B,BL)|T1]):-
get_next_rule_number(R),
gen_fixed(T,T1).
learn_struct_only(DB,R1,R,Score):- %+R1:initial theory of the form [rule(NR,[h],[b]],...], -R:final theory of the same form, -CLL
format("Clause search~n~n",[]),
setting(max_iter,M),
setting(depth_bound,DepthB),
set(depth_bound,false),
findall((H,B,BL),fixed_rule(H,B,BL),LF),
length(LF,LLF),
gen_fixed(LF,LFR),
format("Scoring fixed clauses: ~d clauses~n~n",[LLF]),
score_clause_refinements(LFR,1,LLF,DB,[],NB1,[],CL0,[],CLBG0),
append(NB1,R1,Beam),
cycle_beam(Beam,DB,CL0,[(HCL,S)|TCL],CLBG0,BG,M),
set(depth_bound,DepthB),
format("Theory search~n~n",[]),
setting(max_iter_structure,MS),
cycle_structure(TCL,[HCL],S,-inf,DB,R2,Score,MS),
format("Best target theory~n~n",[]),
write_rules(R2,user_output),
length(BG,NBG),
format("Background search: ~d clauses~n~n",[NBG]),
remove_score(BG,BG2),
append(R2,BG2,R).
learn_struct(DB,R1,R,Score):- %+R1:initial theory of the form [rule(NR,[h],[b]],...], -R:final theory of the same form, -CLL
format("Clause search~n~n",[]),
setting(max_iter,M),
setting(depth_bound,DepthB),
set(depth_bound,false),
findall((H,B,BL),fixed_rule(H,B,BL),LF),
length(LF,LLF),
gen_fixed(LF,LFR),
format("Scoring fixed clauses: ~d clauses~n~n",[LLF]),
score_clause_refinements(LFR,1,LLF,DB,[],NB1,[],CL0,[],CLBG0),
append(NB1,R1,Beam),
cycle_beam(Beam,DB,CL0,[(HCL,S)|TCL],CLBG0,BG,M),
set(depth_bound,DepthB),
format("Theory search~n~n",[]),
setting(max_iter_structure,MS),
cycle_structure(TCL,[HCL],S,-inf,DB,R2,Score,MS),
format("Best target theory~n~n",[]),
write_rules(R2,user_output),
setting(background_clauses,NBG1),
length(BG,NBG),
format("Background search: ~d of ~d clauses~n~n",[NBG1,NBG]),
pick_first(NBG1,BG,BG1),
remove_score(BG,BG2),
write_rules(BG2,user_output),nl,
append(R2,BG1,R).
pick_first(0,_,[]):-!.
pick_first(_,[],[]):-!.
pick_first(N,[(H,_S)|T],[H|T1]):-
N1 is N-1,
pick_first(N1,T,T1).
remove_score([],[]).
remove_score([(H,_S)|T],[H|T1]):-
remove_score(T,T1).
cycle_structure([],R,S,_SP,_DB,R,S,_I):-!. %empty beam
cycle_structure(_CL,R,S,_SP,_DB,R,S,0):-!. %0 iterations
cycle_structure([(RH,_CLL)|RT],R0,S0,SP0,DB,R,S,M):-
already_scored([RH|R0],R3,Score),!,
format("Theory iteration ~d",[M]),nl,nl,
write('Already scored, updated refinement'),nl,
write_rules(R3,user_output),
write('Score '),write(Score),nl,nl,nl,
(Score>S0->
R4=R3,
S4=Score,
SP1=S0
;
R4=R0,
S4=S0,
SP1=SP0
),
M1 is M-1,
cycle_structure(RT,R4,S4,SP1,DB,R,S,M1).
cycle_structure([(RH,_Score)|RT],R0,S0,SP0,DB,R,S,M):-
format("Theory iteration ~d",[M]),nl,nl,
generate_clauses([RH|R0],R2,0,[],Th1),
format("Initial theory~n~n",[]),
write_rules([RH|R0],user_output),
assert_all(Th1),
assert_all(R2),!,
findall(R-HN,(rule(R,HL,_BL,_Lit),length(HL,HN)),L),
keysort(L,LS),
get_heads(LS,LSH),
length(LSH,NR),
init(NR,LSH),
retractall(v(_,_,_)),
length(DB,NEx),
(setting(examples,atoms)->
setting(group,G),
derive_bdd_nodes_groupatoms(DB,NEx,G,[],Nodes,0,CLL0,LE,[]),! % 1 BDD per example if G=1
;
derive_bdd_nodes(DB,NEx,[],Nodes,0,CLL0),! % 1 BDD per model
),
setting(random_restarts_number,N),
format("~nInitial CLL ~f~n~n",[CLL0]),
random_restarts(N,Nodes,CLL0,Score,initial,Par,LE), %output:CLL,Par
format("Score after EMBLEM = ~f~n",[Score]),
retract_all(Th1),
retract_all(R2),!,
end,
update_theory(R2,Par,R3),
write('updated Theory'),nl,
write_rules(R3,user_output), %definite rules without probabilities in the head are not written
(Score>S0->
R4=R3,
S4=Score,
SP1=S0,
write('New best score'),nl
;
R4=R0,
S4=S0,
SP1=SP0
),
store_refinement([RH|R0],R3,Score),
M1 is M-1,
cycle_structure(RT,R4,S4,SP1,DB,R,S,M1).
em(File):-
generate_file_names(File,FileKB,FileIn,FileBG,FileOut,FileL),
reconsult(FileL),
load_models(FileKB,DB),
(file_exists(FileBG)->
set(compiling,on),
load(FileBG,_ThBG,RBG),
set(compiling,off),
generate_clauses(RBG,_RBG1,0,[],ThBG),
assert_all(ThBG)
;
true
),
set(compiling,on),
load(FileIn,_TH,R0),
set(compiling,off),
set(verbosity,3),
statistics(walltime,[_,_]),
learn_params(DB,R0,R,Score),
statistics(walltime,[_,CT]),
CTS is CT/1000,
format("EM: Final score ~f~n",[Score]),
format("Execution time ~f~n~n",[CTS]),
write_rules(R,user_output),
listing(setting/2),
open(FileOut,write,Stream),
format(Stream,'/* EMBLEM Final score ~f~n',[Score]),
format(Stream,'Execution time ~f~n',[CTS]),
tell(Stream),
listing(setting/2),
format(Stream,'*/~n~n',[]),
told,
open(FileOut,append,Stream1),
write_rules(R,Stream1),
close(Stream1).
learn_params(DB,R0,R,Score):- %Parameter Learning
generate_clauses(R0,R1,0,[],Th0),
assert_all(Th0),
assert_all(R1),!,
findall(R-HN,(rule(R,HL,_BL,_Lit),length(HL,HN)),L),
keysort(L,LS),
get_heads(LS,LSH),
length(LSH,NR),
init(NR,LSH),
retractall(v(_,_,_)),
length(DB,NEx),
(setting(examples,atoms)->
setting(group,G),
derive_bdd_nodes_groupatoms(DB,NEx,G,[],Nodes,0,_CLL0,LE,[]),!
;
derive_bdd_nodes(DB,NEx,[],Nodes,0,_CLL0),!
),
setting(random_restarts_number,N),
random_restarts(N,Nodes,-inf,Score,initial,Par,LE), %computes new parameters Par
end,
retract_all(Th0),
retract_all(R1),!,
update_theory_par(R1,Par,R). %replaces in R1 the probabilities Par and outputs R
update_theory_par([],_Par,[]).
update_theory_par([def_rule(H,B,L)|T0],Par,[def_rule(H,B,L)|T]):-!,
update_theory_par(T0,Par,T).
update_theory_par([(H:-B)|T0],Par,[(H:-B)|T]):-!,
update_theory_par(T0,Par,T).
rrr.
update_theory_par([rule(N,_H,_B,_L)|T0],Par,T):-
member([N,[1.0|_T]],Par),!,
rrr,
update_theory_par(T0,Par,T).
update_theory_par([rule(N,H,B,L)|T0],Par,[rule(N,H1,B,L)|T]):-
member([N,P],Par),!,
reverse(P,P1),
update_head_par(H,P1,H1),
update_theory_par(T0,Par,T).
update_theory(R,initial,R):-!.
update_theory([],_Par,[]).
update_theory([def_rule(H,B,L)|T0],Par,[def_rule(H,B,L)|T]):-!,
update_theory(T0,Par,T).
update_theory([(H:-B)|T0],Par,[(H:-B)|T]):-!,
update_theory(T0,Par,T).
update_theory([rule(N,H,B,L)|T0],Par,[rule(N,H1,B,L)|T]):-
member([N,P],Par),!,
reverse(P,P1),
update_head_par(H,P1,H1),
update_theory(T0,Par,T).
update_head_par([],[],[]).
update_head_par([H:_P|T0],[HP|TP],[H:HP|T]):-
update_head_par(T0,TP,T).
cycle_beam([],_DB,CL,CL,CLBG,CLBG,_M):-!.
cycle_beam(_Beam,_DB,CL,CL,CLBG,CLBG,0):-!.
cycle_beam(Beam,DB,CL0,CL,CLBG0,CLBG,M):-
format("Clause iteration ~d",[M]),nl,nl,
cycle_clauses(Beam,DB,[],NB,CL0,CL1,CLBG0,CLBG1),
M1 is M-1,%decreases the number of max_iter M
cycle_beam(NB,DB,CL1,CL,CLBG1,CLBG,M1).
cycle_clauses([],_DB,NB,NB,CL,CL,CLBG,CLBG):-!.
cycle_clauses([(RH,_ScoreH)|T],DB,NB0,NB,CL0,CL,CLBG0,CLBG):-
findall(RS,specialize_rule(RH,RS,_L),LR),!, %-LR:list of lists, each one correponding to a different revised theory; specialize_rule defined in revise.pl
length(LR,NR),
write('Number of revisions '),write(NR),nl,
score_clause_refinements(LR,1,NR,DB,NB0,NB1,CL0,CL1,CLBG0,CLBG1),
cycle_clauses(T,DB,NB1,NB,CL1,CL,CLBG1,CLBG).
score_clause_refinements([],_N,_NR,_DB,NB,NB,CL,CL,CLBG,CLBG).
score_clause_refinements([R1|T],Nrev,NRef,DB,NB0,NB,CL0,CL,CLBG0,CLBG):- %scans the list of revised theories
already_scored_clause(R1,R3,Score),!,
format('Score ref. ~d of ~d~n',[Nrev,NRef]),
write('Already scored, updated refinement'),nl,
write_rules([R3],user_output),
write('Score '),write(Score),nl,nl,nl,
setting(beamsize,BS),
insert_in_order(NB0,(R3,Score),BS,NB1),
Nrev1 is Nrev+1,
score_clause_refinements(T,Nrev1,NRef,DB,NB1,NB,CL0,CL,CLBG0,CLBG).
score_clause_refinements([R1|T],Nrev,NRef,DB,NB0,NB,CL0,CL,CLBG0,CLBG):-
format('Score ref. ~d of ~d~n',[Nrev,NRef]),
write_rules([R1],user_output),
generate_clauses_cw([R1],[R2],0,[],Th1),
assert_all(Th1),
assert_all([R2]),!,
findall(RN-HN,(rule(RN,HL,_BL,_Lit),length(HL,HN)),L),
keysort(L,LS),
get_heads(LS,LSH),
length(LSH,NR),
init(NR,LSH),
retractall(v(_,_,_)),
length(DB,NEx),
get_output_preds(R1,O),
(setting(examples,atoms)->
setting(group,G),
derive_bdd_nodes_groupatoms_output_atoms(DB,O,NEx,G,[],Nodes,0,CLL0,LE,[]),!
;
derive_bdd_nodes(DB,NEx,[],Nodes,0,CLL0),!
),
format("Initial CLL ~f~n",[CLL0]),
setting(random_restarts_REFnumber,N),
random_restarts_ref(N,Nodes,CLL0,Score,initial,Par,LE),
end,
update_theory([R2],Par,[R3]),
write('Updated refinement'),nl,
write_rules([R3],user_output),
write('Score (CLL) '),write(Score),nl,nl,nl,
retract_all(Th1),
retract_all([R2]),!,
setting(beamsize,BS),
insert_in_order(NB0,(R3,Score),BS,NB1),
(target(R3)->
insert_in_order(CL0,(R3,Score),+inf,CL1),
length(CL1,LCL1),
format("N. of target clauses ~d~n~n",[LCL1]),
CLBG1=CLBG0
;
(range_restricted(R3)->
insert_in_order(CLBG0,(R3,Score),+inf,CLBG1),
length(CLBG1,LCL1),
format("N. of background clauses ~d~n~n",[LCL1]),
CL1=CL0
;
format("Not range restricted~n~n",[]),
CL1=CL0,
CLBG1=CLBG0
)
),
store_clause_refinement(R1,R3,Score),
Nrev1 is Nrev+1,
score_clause_refinements(T,Nrev1,NRef,DB,NB1,NB,CL1,CL,CLBG1,CLBG).
range_restricted(rule(_N,HL,BL,_Lit)):-
term_variables(HL,VH),
term_variables(BL,VB),
sublisteq(VH,VB).
sublisteq([],_).
sublisteq([H|T],L):-
member_eq(H,L),
sublisteq(T,L).
target(R):-
get_output_preds(R,O),
member(T,O),
output(T),!.
get_output_preds(rule(_N,HL,_BL,_Lit),O):-
scan_head(HL,[],O).
scan_head(['':_],O,O):-!.
scan_head([],O,O):-!.
scan_head([H:_P|T],O0,O):-
functor(H,F,N),
(member(F/N,O0)->
O1=O0
;
O1=[F/N|O0]
),
scan_head(T,O1,O).
store_clause_refinement(Ref,RefP,Score):-
elab_clause_ref(Ref,Ref1),
recorda(ref_clause,r(Ref1,RefP,Score),_).
store_refinement(Ref,RefP,Score):-
elab_ref(Ref,Ref1),
recorda(ref,r(Ref1,RefP,Score),_).
already_scored_clause(R,R1,Score):-
elab_ref([R],[rule(H,B)]),
recorded(ref_clause,r(rule(H,B1),R1,Score),_),
permutation(B,B1).
already_scored(R,R1,Score):-
elab_ref(R,RR),
recorded(ref,r(RR,R1,Score),_).
elab_clause_ref(rule(_NR,H,B,_Lits),rule(H1,B1)):-
copy_term((H,B),(H1,B1)).
elab_ref([],[]).
elab_ref([rule(_NR,H,B,_Lits)|T],[rule(H1,B1)|T1]):-
copy_term((H,B),(H1,B1)),
numbervars((H1,B1),0,_N),
elab_ref(T,T1).
elab_ref([def_rule(H,B,_Lits)|T],[rule(H1,B1)|T1]):-
copy_term((H,B),(H1,B1)),
numbervars((H1,B1),0,_N),
elab_ref(T,T1).
%insertion in the beam
insert_in_order([],C,BeamSize,[C]):-
BeamSize>0,!.
insert_in_order(Beam,_New,0,Beam):-!.
insert_in_order([(Th1,Heuristic1)|RestBeamIn],(Th,Heuristic),BeamSize,BeamOut):-
Heuristic>Heuristic1,!,
% larger heuristic, insert here
NewBeam=[(Th,Heuristic),(Th1,Heuristic1)|RestBeamIn],
length(NewBeam,L),
(L>BeamSize->
nth(L,NewBeam,_Last,BeamOut)
;
BeamOut=NewBeam
).
insert_in_order([(Th1,Heuristic1)|RestBeamIn],(Th,Heuristic),BeamSize,
[(Th1,Heuristic1)|RestBeamOut]):-
BeamSize1 is BeamSize -1,
insert_in_order(RestBeamIn,(Th,Heuristic),BeamSize1,
RestBeamOut).
remove_int_atom_list([],[]).
remove_int_atom_list([A|T],[A1|T1]):-
A=..[F,_|Arg],
A1=..[F|Arg],
remove_int_atom_list(T,T1).
remove_int_atom(A,A1):-
A=..[F,_|T],
A1=..[F|T].
get_heads([],[]).
get_heads([_-H|T],[H|TN]):-
get_heads(T,TN).
derive_bdd_nodes([],_E,Nodes,Nodes,CLL,CLL).
derive_bdd_nodes([H|T],E,Nodes0,Nodes,CLL0,CLL):-
get_output_atoms(O),
generate_goal(O,H,[],GL),
(prob(H,P)->
CardEx is P*E
;
CardEx is 1.0
),
init_bdd,
one(One),
get_node_list(GL,One,BDD,CardEx),
ret_prob(BDD,HP),
(HP=:=0.0->
setting(logzero,LZ),
CLL1 is CLL0+LZ*CardEx
;
CLL1 is CLL0+log(HP)*CardEx
),
end_bdd,
append(Nodes0,[[BDD,CardEx]],Nodes1),
derive_bdd_nodes(T,E,Nodes1,Nodes,CLL1,CLL).
get_node_list([],BDD,BDD,_CE).
get_node_list([H|T],BDD0,BDD,CE):-
get_node(H,BDD1),
and(BDD0,BDD1,BDD2),
get_node_list(T,BDD2,BDD,CE).
derive_bdd_nodes_groupatoms_output_atoms([],_O,_E,_G,Nodes,Nodes,CLL,CLL,LE,LE).
derive_bdd_nodes_groupatoms_output_atoms([H|T],O,E,G,Nodes0,Nodes,CLL0,CLL,LE0,LE):-
generate_goal(O,H,[],GL),
length(GL,NA),
(prob(H,P)->
CardEx is P*E/NA
;
CardEx is 1.0
),
get_node_list_groupatoms(GL,BDDs,CardEx,G,CLL0,CLL1,LE0,LE1),
append(Nodes0,BDDs,Nodes1),
derive_bdd_nodes_groupatoms_output_atoms(T,O,E,G,Nodes1,Nodes,CLL1,CLL,LE1,LE).
derive_bdd_nodes_groupatoms([],_E,_G,Nodes,Nodes,CLL,CLL,LE,LE).
derive_bdd_nodes_groupatoms([H|T],E,G,Nodes0,Nodes,CLL0,CLL,LE0,LE):-
get_output_atoms(O),
generate_goal(O,H,[],GL),
length(GL,NA),
(prob(H,P)->
CardEx is P*E/NA
;
CardEx is 1.0
),
get_node_list_groupatoms(GL,BDDs,CardEx,G,CLL0,CLL1,LE0,LE1),
append(Nodes0,BDDs,Nodes1),
derive_bdd_nodes_groupatoms(T,E,G,Nodes1,Nodes,CLL1,CLL,LE1,LE).
get_node_list_groupatoms([],[],_CE,_Gmax,CLL,CLL,LE,LE).
get_node_list_groupatoms([H|T],[[BDD,CE1]|BDDT],CE,Gmax,CLL0,CLL,LE0,LE):-
init_bdd,
one(One),
get_bdd_group([H|T],T1,Gmax,G,One,BDD,CE,LE0,LE1), %output:BDD,CLL
CE1 is CE*(Gmax-G),
ret_prob(BDD,HP),
end_bdd,
(HP =:=0.0->
setting(logzero,LZ),
CLL2 is CLL0+LZ*CE1
;
CLL2 is CLL0+log(HP)*CE1
),
get_node_list_groupatoms(T1,BDDT,CE,Gmax,CLL2,CLL,LE1,LE).
get_bdd_group([],[],G,G,BDD,BDD,_CE,LE,LE):-!.
get_bdd_group(T,T,0,0,BDD,BDD,_CE,LE,LE):- !.
get_bdd_group([H|T],T1,Gmax,G1,BDD0,BDD,CE,[H|LE0],LE):-
% write(H),flush_output,
get_node(H,BDD1), %creates the BDD for atom H
and(BDD0,BDD1,BDD2),
G is Gmax-1,
get_bdd_group(T,T1,G,G1,BDD2,BDD,CE,LE0,LE).
/* EM start */
random_restarts(0,_Nodes,Score,Score,Par,Par,_LE):-!.
random_restarts(N,Nodes,Score0,Score,Par0,Par,LE):-
setting(verbosity,Ver),
(Ver>2->
setting(random_restarts_number,NMax),
Num is NMax-N+1,
format("Restart number ~d~n~n",[Num]),
flush_output
;
true
),
randomize,
setting(epsilon_em,EA),
setting(epsilon_em_fraction,ER),
length(Nodes,L),
setting(iter,Iter),
em(Nodes,EA,ER,L,Iter,CLL,Par1,ExP),
score(LE,ExP,CLL,ScoreR),
setting(verbosity,Ver),
(Ver>2->
format("Random_restart: Score ~f~n",[ScoreR])
;
true
),
N1 is N-1,
(ScoreR>Score0->
random_restarts(N1,Nodes,ScoreR,Score,Par1,Par,LE)
;
random_restarts(N1,Nodes,Score0,Score,Par0,Par,LE)
).
random_restarts_ref(0,_Nodes,Score,Score,Par,Par,_LE):-!.
random_restarts_ref(N,Nodes,Score0,Score,Par0,Par,LE):-
setting(verbosity,Ver),
(Ver>2->
setting(random_restarts_REFnumber,NMax),
Num is NMax-N+1,
format("Restart number ~d~n~n",[Num]),
flush_output
;
true
),
setting(epsilon_em,EA),
setting(epsilon_em_fraction,ER),
length(Nodes,L),
setting(iterREF,Iter),
em(Nodes,EA,ER,L,Iter,CLLR,Par1,ExP),
score(LE,ExP,CLLR,ScoreR),
setting(verbosity,Ver),
(Ver>2->
format("Random_restart: Score ~f~n",[ScoreR])
;
true
),
N1 is N-1,
(ScoreR>Score0->
random_restarts_ref(N1,Nodes,ScoreR,Score,Par1,Par,LE)
;
random_restarts_ref(N1,Nodes,Score0,Score,Par0,Par,LE)
).
score(_LE,_ExP,CLL,CLL):-
setting(score,ll),!.
score(LE,ExP,_CLL,Score):-
compute_prob(LE,ExP,LPU,0,Pos,0,Neg),
keysort(LPU,LPO),
reverse(LPO,LP),
compute_aucpr(LP,Pos,Neg,Score).
compute_prob([],[],[],Pos,Pos,Neg,Neg).
compute_prob([\+ HE|TE],[HP|TP],[P- (\+ HE)|T],Pos0,Pos,Neg0,Neg):-!,
P is 1-HP,
Neg1 is Neg0+1,
compute_prob(TE,TP,T,Pos0,Pos,Neg1,Neg).
compute_prob([ HE|TE],[HP|TP],[HP- HE|T],Pos0,Pos,Neg0,Neg):-
Pos1 is Pos0+1,
compute_prob(TE,TP,T,Pos1,Pos,Neg0,Neg).
compute_aucpr(L,Pos,Neg,A):-
L=[P_0-E|TL],
(E= (\+ _ )->
FP=1,
TP=0,
FN=Pos,
TN is Neg -1
;
FP=0,
TP=1,
FN is Pos -1,
TN=Neg
),
compute_curve_points(TL,P_0,TP,FP,FN,TN,Points),
Points=[R0-P0|_TPoints],
(R0=:=0,P0=:=0->
Flag=true
;
Flag=false
),
area(Points,Flag,Pos,0,0,0,A).
compute_curve_points([],_P0,TP,FP,_FN,_TN,[1.0-Prec]):-!,
Prec is TP/(TP+FP).
compute_curve_points([P- (\+ _)|T],P0,TP,FP,FN,TN,Pr):-!,
(P<P0->
Prec is TP/(TP+FP),
Rec is TP/(TP+FN),
Pr=[Rec-Prec|Pr1],
P1=P
;
Pr=Pr1,
P1=P0
),
FP1 is FP+1,
TN1 is TN-1,
compute_curve_points(T,P1,TP,FP1,FN,TN1,Pr1).
compute_curve_points([P- _|T],P0,TP,FP,FN,TN,Pr):-!,
(P<P0->
Prec is TP/(TP+FP),
Rec is TP/(TP+FN),
Pr=[Rec-Prec|Pr1],
P1=P
;
Pr=Pr1,
P1=P0
),
TP1 is TP+1,
FN1 is FN-1,
compute_curve_points(T,P1,TP1,FP,FN1,TN,Pr1).
area([],_Flag,_Pos,_TPA,_FPA,A,A).
area([R0-P0|T],Flag,Pos,TPA,FPA,A0,A):-
TPB is R0*Pos,
(TPB=:=0->
A1=A0,
FPB=0
;
R_1 is TPA/Pos,
(TPA=:=0->
(Flag=false->
P_1=P0
;
P_1=0.0
)
;
P_1 is TPA/(TPA+FPA)
),
FPB is TPB*(1-P0)/P0,
N is TPB-TPA+0.5,
interpolate(1,N,Pos,R_1,P_1,TPA,FPA,TPB,FPB,A0,A1)
),
area(T,Flag,Pos,TPB,FPB,A1,A).
interpolate(I,N,_Pos,_R0,_P0,_TPA,_FPA,_TPB,_FPB,A,A):-I>N,!.
interpolate(I,N,Pos,R0,P0,TPA,FPA,TPB,FPB,A0,A):-
R is (TPA+I)/Pos,
P is (TPA+I)/(TPA+I+FPA+(FPB-FPA)/(TPB-TPA)*I),
A1 is A0+(R-R0)*(P+P0)/2,
I1 is I+1,
interpolate(I1,N,Pos,R,P,TPA,FPA,TPB,FPB,A1,A).
randomize([],[]):-!.
randomize([rule(N,V,NH,HL,BL,LogF)|T],[rule(N,V,NH,HL1,BL,LogF)|T1]):-
length(HL,L),
Int is 1.0/L,
randomize_head(Int,HL,0,HL1),
randomize(T,T1).
randomize_head(_Int,['':_],P,['':PNull1]):-!,
PNull is 1.0-P,
(PNull>=0.0->
PNull1 =PNull
;
PNull1=0.0
).
randomize_head(Int,[H:_|T],P,[H:PH1|NT]):-
PMax is 1.0-P,
random(0,PMax,PH1),
P1 is P+PH1,
randomize_head(Int,T,P1,NT).
update_head([],[],_N,[]):-!.
update_head([H:_P|T],[PU|TP],N,[H:P|T1]):-
P is PU/N,
update_head(T,TP,N,T1).
/* EM end */
/* utilities */
generate_file_names(File,FileKB,FileIn,FileBG,FileOut,FileL):-
generate_file_name(File,".kb",FileKB),
generate_file_name(File,".cpl",FileIn),
generate_file_name(File,".rules",FileOut),
generate_file_name(File,".bg",FileBG),
generate_file_name(File,".l",FileL).
generate_file_name(File,Ext,FileExt):-
name(File,FileString),
append(FileString,Ext,FileStringExt),
name(FileExt,FileStringExt).
load_models(File,ModulesList):- %carica le interpretazioni, 1 alla volta
open(File,read,Stream),
read_models(Stream,ModulesList),
close(Stream).
read_models(Stream,[Name1|Names]):-
read(Stream,begin(model(Name))),!,
(number(Name)->
name(Name,NameStr),
append("i",NameStr,Name1Str),
name(Name1,Name1Str)
;
Name1=Name
),
read_all_atoms(Stream,Name1),
read_models(Stream,Names).
read_models(_S,[]).
read_all_atoms(Stream,Name):-
read(Stream,At),
At \=end(model(_Name)),!,
(At=neg(Atom)->
Atom=..[Pred|Args],
Atom1=..[Pred,Name|Args],
assertz(neg(Atom1))
;
(At=prob(Pr)->
assertz(prob(Name,Pr))
;
At=..[Pred|Args],
Atom1=..[Pred,Name|Args],
assertz(Atom1)
)
),
read_all_atoms(Stream,Name).
read_all_atoms(_S,_N).
write_param(initial,S):-!,
format("~nInitial parameters~n",[]),
findall(rule(R,H,B,Lit),rule(R,H,B,Lit),LDis),
findall(rule(d,[H:1.0],B,Lit),def_rule(H,B,Lit),LDef),
append(LDis,LDef,L),
write_model(L,S).
write_param(L,S):-
reverse(L,L1),
write_par(L1,S).
write_par([],S):-
findall(rule(d,[H:1.0],B,Lit),def_rule(H,B,Lit),L),
write_model(L,S).
write_par([[N,P]|T],S):-
rule(N,HL0,BL),
reverse(P,PR),
new_par(PR,HL0,HL),
copy_term((HL,BL),(HL1,BL1)),
numbervars((HL1,BL1),0,_M),
write_disj_clause(S,(HL1:-BL1)),
write_par(T,S).
write_rules([],_S).
write_rules([rule(_N,HL,BL,Lit)|T],S):-
copy_term((HL,BL,Lit),(HL1,BL1,Lit1)),
numbervars((HL1,BL1,Lit1),0,_M),
write_disj_clause(S,(HL1:-BL1)),
% write(Lit1),nl,
write_rules(T,S).
new_par([],[],[]).
new_par([HP|TP],[Head:_|TO],[Head:HP|TN]):-
new_par(TP,TO,TN).
write_model([],_Stream):-!.
write_model([rule(_N,HL,BL)|Rest],Stream):-
copy_term((HL,BL),(HL1,BL1)),
numbervars((HL1,BL1),0,_M),
write_disj_clause(Stream,(HL1:-BL1)),
write_model(Rest,Stream).
write_disj_clause(S,(H:-[])):-!,
write_head(S,H),
format(S,".~n~n",[]).
write_disj_clause(S,(H:-B)):-
write_head(S,H),
write(S,' :-'),
nl(S),
write_body(S,B).
write_head(S,[A:1.0|_Rest]):-!,
% write(_Rest),nl,
format(S,"~p:1.0",[A]).
write_head(S,[A:P,'':_P]):-!,
format(S,"~p:~g",[A,P]).
write_head(S,[A:P]):-!,
format(S,"~p:~g",[A,P]).
write_head(S,[A:P|Rest]):-
format(S,"~p:~g ; ",[A,P]),
write_head(S,Rest).
write_body(S,[]):-
format(S,"\ttrue.~n~n",[]).
write_body(S,[A]):-!,
format(S,"\t~p.~n~n",[A]).
write_body(S,[A|T]):-
format(S,"\t~p,~n",[A]),
write_body(S,T).
list2or([],true):-!.
list2or([X],X):-
X\=;(_,_),!.
list2or([H|T],(H ; Ta)):-!,
list2or(T,Ta).
list2and([],true):-!.
list2and([X],X):-
X\=(_,_),!.
list2and([H|T],(H,Ta)):-!,
list2and(T,Ta).
deduct(0,_DB,Th,Th):-!.
deduct(NM,DB,InTheory0,InTheory):-
get_head_atoms(O),
sample(1,DB,[M],DB1),
generate_head(O,M,[],HL),
generate_body(HL,InTheory1),
append(InTheory0,InTheory1,InTheory2),
NM1 is NM-1,
deduct(NM1,DB1,InTheory2,InTheory).
get_head_atoms(O):-
findall(A,modeh(_,A),O0),
findall((A,B,D),modeh(_,A,B,D),O1),
append(O0,O1,O).
get_head_atoms(LH,LH0):-
setof(P/Ar,head_predicate(P/Ar),LP),
scan_pred_list(LP,LH,LH0).
scan_pred_list([],[],[]).
scan_pred_list([P/Ar|T],[(P/Ar,LHP)|LH],[(P/Ar,[])|T1]):-
output(P/Ar),!,
findall(A,C^(modeh(C,A),functor(A,P,Ar)),LHS),
findall((A,B,Cons,D),(C,At)^(modeh(C,A,B,Cons,D),member(At,A),functor(At,P,Ar)),LHC),
append(LHS,LHC,LHP),
scan_pred_list(T,LH,T1).
scan_pred_list([P/Ar|T],[(P/Ar,LHP)|LH],[(P/Ar,[])|T1]):-
findall(A,C^(modeh(C,A),functor(A,P,Ar)),LHS),
findall((A,B,Cons,D),(C,At,At1,P1,Ar1)^(
modeh(C,A,B,Cons,D),member(At,A),functor(At,P,Ar),
member(At1,A),functor(At1,P1,Ar1),\+ output(P1/Ar1)),LHC),
append(LHS,LHC,LHP),
scan_pred_list(T,LH,T1).
head_predicate(P/Ar):-
modeh(_C,A),
functor(A,P,Ar).
head_predicate(P/Ar):-
modeh(_C,A,_B,_Cons,_D),
member(At,A),
functor(At,P,Ar).
generate_top_cl([],[]):-!.
generate_top_cl([(P/A,LMH)|T],[(P/A,LR)|TR]):-
generate_top_cl_pred(LMH,LR),
generate_top_cl(T,TR).
generate_top_cl_pred([],[]):-!.
generate_top_cl_pred([A|T],[(rule(R,[A1:0.5,'':0.5],[],true),-inf)|TR]):-
A=..[F|ArgM],
keep_const(ArgM,Arg),
A1=..[F|Arg],
get_next_rule_number(R),
generate_top_cl_pred(T,TR).
generate_head(0,_DB,_LMH,LH,LH):-!.
generate_head(NM,DB,LMH,LH0,LH):-
sample(1,DB,[M],DB1),
generate_head_ex(LMH,M,LH0,LH1),
NM1 is NM-1,
generate_head(NM1,DB1,LMH,LH1,LH).
generate_head_ex([],_M,[],[]).
generate_head_ex([(P/A,L)|T],M,[(P/A,LH)|LHT],[(P/A,LH1)|LHT1]):-
generate_head_pred(L,M,[],LHP),
append(LH,LHP,LH1),
generate_head_ex(T,M,LHT,LHT1).
generate_head_pred([],_M,HL,HL):-!.
generate_head_pred([(A,G,Cons,D)|T],M,H0,H1):-!,
generate_head_goal(G,M,Goals),
findall((A,Goals,D),(member(Goal,Goals),call(Goal),call(Cons),ground(Goals)),L),
setting(initial_clauses_per_megaex,IC), %IC: represents how many samples are extracted from the list L of example
sample(IC,L,L1), %+IC,L, -L1
append(H0,L1,H2),
generate_head_pred(T,M,H2,H1).
generate_head_pred([A|T],M,H0,H1):-
functor(A,F,N),
functor(F1,F,N),
F1=..[F|Arg],
Pred1=..[F,M|Arg],
A=..[F|ArgM],
keep_const(ArgM,Arg),
findall((A,Pred1),call(Pred1),L),
setting(initial_clauses_per_megaex,IC), %IC: represents how many samples are extracted from the list L of example
sample(IC,L,L1), %+IC,L, -L1
append(H0,L1,H2),
generate_head_pred(T,M,H2,H1).
generate_head_goal([],_M,[]).
generate_head_goal([H|T],M,[H1|T1]):-
H=..[F|Arg],
H1=..[F,M|Arg],
generate_head_goal(T,M,T1).
keep_const([],[]).
keep_const([- _|T],[_|T1]):-!,
keep_const(T,T1).
keep_const([+ _|T],[_|T1]):-!,
keep_const(T,T1).
keep_const([-# _|T],[_|T1]):-!,
keep_const(T,T1).
keep_const([H|T],[H|T1]):-
keep_const(T,T1).
sample(0,List,[],List):-!.
sample(N,List,List,[]):-
length(List,L),
L=<N,!.
sample(N,List,[El|List1],Li):-
length(List,L),
random(0,L,Pos),
nth0(Pos,List,El,Rest),
N1 is N-1,
sample(N1,Rest,List1,Li).
sample(0,_List,[]):-!.
sample(N,List,List):-
length(List,L),
L=<N,!.
sample(N,List,[El|List1]):-
length(List,L),
random(0,L,Pos),
nth0(Pos,List,El,Rest),
N1 is N-1,
sample(N1,Rest,List1).
get_args([],[],[],A,A,AT,AT,_).
get_args([HM|TM],[H|TH],[(H,HM)|TP],A0,A,AT0,AT,M):-
HM=..[F|ArgsTypes],
H=..[F,M|Args],
append(A0,Args,A1),
append(AT0,ArgsTypes,AT1),
get_args(TM,TH,TP,A1,A,AT1,AT,M).
/* Generation of the bottom clauses */
gen_head([],P,['':P]).
gen_head([H|T],P,[H:P|TH]):-
gen_head(T,P,TH).
get_modeb([],B,B).
get_modeb([F/AA|T],B0,B):-
findall((R,B),(modeb(R,B),functor(B,F,AA)),BL),
append(B0,BL,B1),
get_modeb(T,B1,B).
generate_body([],[]).
generate_body([(P/A,LH)|T],[(P/A,LR)|TR]):-
generate_body_pred(LH,LR),
generate_body(T,TR).
generate_body_pred([],[]):-!.
generate_body_pred([(A,H,Det)|T],[(rule(R,HP,[],BodyList),-inf)|CL0]):-!,
get_modeb(Det,[],BL),
get_args(A,H,Pairs,[],Args,[],ArgsTypes,M),
setting(d,D),
cycle_modeb(ArgsTypes,Args,[],[],BL,a,[],BLout0,D,M),
remove_duplicates(BLout0,BLout),
variabilize((Pairs:-BLout),CLV), %+(Head):-Bodylist; -CLV:(Head):-Bodylist with variables _num in place of constants
CLV=(Head1:-BodyList1),
remove_int_atom_list(Head1,Head),
remove_int_atom_list(BodyList1,BodyList),
get_next_rule_number(R),
length(Head,LH),
Prob is 1/(LH+1),
gen_head(Head,Prob,HP),
copy_term((HP,BodyList),(HeadV,BodyListV)),
numbervars((HeadV,BodyListV),0,_V),
format("Bottom clause: example ~p~nClause~n",[H]),
write_disj_clause(user_output,(HeadV:-BodyListV)),
generate_body_pred(T,CL0).
generate_body_pred([(A,H)|T],[(rule(R,[Head:0.5,'':0.5],[],BodyList),-inf)|CL0]):-
functor(A,F,AA),
findall((R,B),(modeb(R,B),functor(B,FB,AB),determination(F/AA,FB/AB)),BL),
A=..[F|ArgsTypes],
H=..[F,M|Args],
setting(d,D),
cycle_modeb(ArgsTypes,Args,[],[],BL,a,[],BLout0,D,M),
remove_duplicates(BLout0,BLout),
variabilize(([(H,A)]:-BLout),CLV), %+(Head):-Bodylist; -CLV:(Head):-Bodylist with variables _num in place of constants
CLV=([Head1]:-BodyList1),
remove_int_atom(Head1,Head),
remove_int_atom_list(BodyList1,BodyList),
get_next_rule_number(R),
copy_term((Head,BodyList),(HeadV,BodyListV)),
numbervars((HeadV,BodyListV),0,_V),
format("Bottom clause: example ~p~nClause~n~p:0.5 :-~n",[H,HeadV]),
write_body(user_output,BodyListV),
generate_body_pred(T,CL0).
variabilize((H:-B),(H1:-B1)):-
variabilize_list(H,H1,[],AS,M),
variabilize_list(B,B1,AS,_AS,M).
variabilize_list([],[],A,A,_M).
variabilize_list([(H,Mode)|T],[H1|T1],A0,A,M):-
builtin(H),!,
H=..[F|Args],
Mode=..[F|ArgTypes],
variabilize_args(Args,ArgTypes, Args1,A0,A1),
H1=..[F,M|Args1],
variabilize_list(T,T1,A1,A,M).
variabilize_list([(H,Mode)|T],[H1|T1],A0,A,M):-
H=..[F,_M|Args],
Mode=..[F|ArgTypes],
variabilize_args(Args,ArgTypes, Args1,A0,A1),
H1=..[F,M|Args1],
variabilize_list(T,T1,A1,A,M).
variabilize_args([],[],[],A,A).
variabilize_args([C|T],[C|TT],[C|TV],A0,A):-!,
variabilize_args(T,TT,TV,A0,A).
variabilize_args([C|T],[# _Ty|TT],[C|TV],A0,A):-!,
variabilize_args(T,TT,TV,A0,A).
variabilize_args([C|T],[-# _Ty|TT],[C|TV],A0,A):-!,
variabilize_args(T,TT,TV,A0,A).
variabilize_args([C|T],[_Ty|TT],[V|TV],A0,A):-
member(C/V,A0),!,
variabilize_args(T,TT,TV,A0,A).
variabilize_args([C|T],[_Ty|TT],[V|TV],A0,A):-
variabilize_args(T,TT,TV,[C/V|A0],A).
cycle_modeb(ArgsTypes,Args,ArgsTypes,Args,_BL,L,L,L,_,_M):-!.
cycle_modeb(_ArgsTypes,_Args,_ArgsTypes1,_Args1,_BL,_L,L,L,0,_M):-!.
cycle_modeb(ArgsTypes,Args,_ArgsTypes0,_Args0,BL,_L0,L1,L,D,M):-
find_atoms(BL,ArgsTypes,Args,ArgsTypes1,Args1,L1,L2,M),
D1 is D-1,
cycle_modeb(ArgsTypes1,Args1,ArgsTypes,Args,BL,L1,L2,L,D1,M).
find_atoms([],ArgsTypes,Args,ArgsTypes,Args,L,L,_M).
find_atoms([(R,H)|T],ArgsTypes0,Args0,ArgsTypes,Args,L0,L1,M):-
H=..[F|ArgsT],
findall((A,H),instantiate_query(ArgsT,ArgsTypes0,Args0,F,M,A),L),
call_atoms(L,[],At),
remove_duplicates(At,At1),
(R = '*' ->
R1= +inf
;
R1=R
),
sample(R1,At1,At2),
extract_output_args(At2,ArgsT,ArgsTypes0,Args0,ArgsTypes1,Args1),
append(L0,At2,L2),
find_atoms(T,ArgsTypes1,Args1,ArgsTypes,Args,L2,L1,M).
call_atoms([],A,A).
call_atoms([(H,M)|T],A0,A):-
findall((H,M),H,L),
append(A0,L,A1),
call_atoms(T,A1,A).
extract_output_args([],_ArgsT,ArgsTypes,Args,ArgsTypes,Args).
extract_output_args([(H,_At)|T],ArgsT,ArgsTypes0,Args0,ArgsTypes,Args):-
builtin(H),!,
H=..[_F|ArgsH],
add_const(ArgsH,ArgsT,ArgsTypes0,Args0,ArgsTypes1,Args1),
extract_output_args(T,ArgsT,ArgsTypes1,Args1,ArgsTypes,Args).
extract_output_args([(H,_At)|T],ArgsT,ArgsTypes0,Args0,ArgsTypes,Args):-
H=..[_F,_M|ArgsH],
add_const(ArgsH,ArgsT,ArgsTypes0,Args0,ArgsTypes1,Args1),
extract_output_args(T,ArgsT,ArgsTypes1,Args1,ArgsTypes,Args).
add_const([],[],ArgsTypes,Args,ArgsTypes,Args).
add_const([_A|T],[+_T|TT],ArgsTypes0,Args0,ArgsTypes,Args):-!,
add_const(T,TT,ArgsTypes0,Args0,ArgsTypes,Args).
add_const([A|T],[-Type|TT],ArgsTypes0,Args0,ArgsTypes,Args):-!,
(already_present(ArgsTypes0,Args0,A,Type)->
ArgsTypes1=ArgsTypes0,
Args1=Args0
;
ArgsTypes1=[+Type|ArgsTypes0],
Args1=[A|Args0]
),
add_const(T,TT,ArgsTypes1,Args1,ArgsTypes,Args).
add_const([A|T],[-# Type|TT],ArgsTypes0,Args0,ArgsTypes,Args):-!,
(already_present(ArgsTypes0,Args0,A,Type)->
ArgsTypes1=ArgsTypes0,
Args1=Args0
;
ArgsTypes1=[+Type|ArgsTypes0],
Args1=[A|Args0]
),
add_const(T,TT,ArgsTypes1,Args1,ArgsTypes,Args).
add_const([_A|T],[# _|TT],ArgsTypes0,Args0,ArgsTypes,Args):-!,
add_const(T,TT,ArgsTypes0,Args0,ArgsTypes,Args).
add_const([A|T],[A|TT],ArgsTypes0,Args0,ArgsTypes,Args):-
add_const(T,TT,ArgsTypes0,Args0,ArgsTypes,Args).
already_present([+T|_TT],[C|_TC],C,T):-!.
already_present([_|TT],[_|TC],C,T):-
already_present(TT,TC,C,T).
instantiate_query(ArgsT,ArgsTypes,Args,F,M,A):-
instantiate_input(ArgsT,ArgsTypes,Args,ArgsB),
A1=..[F|ArgsB],
(builtin(A1)->
A=A1
;
A=..[F,M|ArgsB]
).
instantiate_input([],_AT,_A,[]).
instantiate_input([-_Type|T],AT,A,[_V|TA]):-!,
instantiate_input(T,AT,A,TA).
instantiate_input([+Type|T],AT,A,[H|TA]):-!,
find_val(AT,A,+Type,H),
instantiate_input(T,AT,A,TA).
instantiate_input([# Type|T],AT,A,[H|TA]):-!,
find_val(AT,A,+Type,H),
instantiate_input(T,AT,A,TA).
instantiate_input([-# _Type|T],AT,A,[_V|TA]):-!,
instantiate_input(T,AT,A,TA).
instantiate_input([C|T],AT,A,[C|TA]):-
instantiate_input(T,AT,A,TA).
find_val([T|_TT],[A|_TA],T,A).
find_val([_T|TT],[_A|TA],T,A):-
find_val(TT,TA,T,A).
get_output_atoms(O):-
findall((A/Ar),output((A/Ar)),O).
generate_goal([],_H,G,G):-!.
generate_goal([P/A|T],H,G0,G1):-
functor(Pred,P,A),
Pred=..[P|Rest],
Pred1=..[P,H|Rest],
findall(Pred1,call(Pred1),L),
findall(\+ Pred1,call(neg(Pred1)),LN),
append(G0,L,G2),
append(G2,LN,G3),
generate_goal(T,H,G3,G1).
:-[inference_lemur].