This repository has been archived on 2023-08-20.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/CLPBN/clpbn/bdd.yap

1067 lines
31 KiB
Prolog

/************************************************
BDDs in CLP(BN)
A variable is represented by the N possible cases it can take
V = v(Va, Vb, Vc)
The generic formula is
V <- X, Y
Va <- P*X1*Y1 + Q*X2*Y2 + ...
**************************************************/
:- module(clpbn_bdd,
[bdd/3,
set_solver_parameter/2,
init_bdd_solver/4,
init_bdd_ground_solver/5,
run_bdd_solver/3,
run_bdd_ground_solver/3,
finalize_bdd_solver/1,
check_if_bdd_done/1,
call_bdd_ground_solver/6
]).
:- use_module(library('clpbn/dists'),
[dist/4,
get_dist_domain/2,
get_dist_domain_size/2,
get_dist_all_sizes/2,
get_dist_params/2
]).
:- use_module(library('clpbn/display'),
[clpbn_bind_vals/3]).
:- use_module(library('clpbn/aggregates'),
[check_for_agg_vars/2]).
:- use_module(library(atts)).
:- use_module(library(hacks)).
:- use_module(library(lists)).
:- use_module(library(dgraphs)).
:- use_module(library(bdd)).
:- use_module(library(ddnnf)).
:- use_module(library(simpbool)).
:- use_module(library(rbtrees)).
:- use_module(library(bhash)).
:- use_module(library(matrix)).
:- use_module(library(maplist)).
:- use_module(library(clpbn/numbers)).
:- dynamic network_counting/1.
:- attribute order/1.
:- dynamic bdds/1.
%bdds(ddnnf).
bdds(bdd).
%
% QVars: all query variables?
%
%
init_bdd_ground_solver(QueryKeys, AllKeys, Factors, Evidence, bdd(QueryKeys, AllKeys, Factors, Evidence)).
%
% just call horus solver.
%
run_bdd_ground_solver(_QueryVars, Solutions, bdd(GKeys, Keys, Factors, Evidence) ) :- !,
call_bdd_ground_solver_for_probabilities(GKeys, Keys, Factors, Evidence, Solutions).
check_if_bdd_done(_Var).
call_bdd_ground_solver(QueryVars, QueryKeys, AllKeys, Factors, Evidence, Output) :-
call_bdd_ground_solver_for_probabilities([QueryKeys], AllKeys, Factors, Evidence, Solutions),
clpbn_bind_vals([QueryVars], Solutions, Output).
call_bdd_ground_solver_for_probabilities(QueryKeys, AllKeys, Factors, Evidence, Solutions) :-
keys_to_numbers(AllKeys, Factors, Evidence, Hash4, Id4, FactorIds, EvidenceIds),
init_bdd(FactorIds, EvidenceIds, Hash4, Id4, BDD),
run_solver(QueryKeys, Solutions, BDD).
init_bdd(FactorIds, EvidenceIds, Hash, Id, bdd(Term, Leaves, Tops, Hash, Id)) :-
sort_keys(FactorIds, AllVars, Leaves),
rb_new(OrderVs0),
foldl2(order_key, AllVars, 0, _, OrderVs0, OrderVs),
rb_new(Vars0),
rb_new(Pars0),
rb_new(Ev0),
foldl(evtotree,EvidenceIds,Ev0,Ev),
rb_new(Fs0),
foldl(ftotree,FactorIds,Fs0,Fs),
init_tops(Leaves,Tops),
get_keys_info(AllVars, Ev, Fs, OrderVs, Vars0, _Vars, Pars0, _Pars, Leaves, Tops, Term, []).
order_key( Id, I0, I, OrderVs0, OrderVs) :-
I is I0+1,
rb_insert(OrderVs0, Id, I0, OrderVs).
evtotree(K=V,Ev0,Ev) :-
rb_insert(Ev0, K, V, Ev).
ftotree(F, Fs0, Fs) :-
F = fn([K|_Parents],_,_,_,_),
rb_insert(Fs0, K, F, Fs).
bdd([[]],_,_) :- !.
bdd([QueryVars], AllVars, AllDiffs) :-
init_bdd_solver(_, AllVars, _, BayesNet),
run_bdd_solver([QueryVars], LPs, BayesNet),
finalize_bdd_solver(BayesNet),
clpbn_bind_vals([QueryVars], [LPs], AllDiffs).
init_bdd_solver(_, AllVars0, _, bdd(Term, Leaves, Tops)) :-
% check_for_agg_vars(AllVars0, AllVars1),
AllVars0 = AllVars1,
sort_vars(AllVars1, AllVars, Leaves),
order_vars(AllVars, 0),
rb_new(Vars0),
rb_new(Pars0),
init_tops(Leaves,Tops),
get_vars_info(AllVars, Vars0, _Vars, Pars0, _Pars, Leaves, Tops, Term, []).
order_vars([], _).
order_vars([V|AllVars], I0) :-
put_atts(V, [order(I0)]),
I is I0+1,
order_vars(AllVars, I).
init_tops([],[]).
init_tops([_|Leaves],[_|Tops]) :-
init_tops(Leaves,Tops).
sort_keys(AllFs, AllVars, Leaves) :-
dgraph_new(Graph0),
foldl(add_node, AllFs, Graph0, Graph),
dgraph_leaves(Graph, Leaves),
dgraph_top_sort(Graph, AllVars).
add_node(fn([K|Parents],_,_,_,_), Graph0, Graph) :-
dgraph_add_vertex(Graph0, K, Graph1),
foldl(add_edge(K), Parents, Graph1, Graph).
add_edge(K, K0, Graph0, Graph) :-
dgraph_add_edge(Graph0, K0, K, Graph).
sort_vars(AllVars0, AllVars, Leaves) :-
dgraph_new(Graph0),
build_graph(AllVars0, Graph0, Graph),
dgraph_leaves(Graph, Leaves),
dgraph_top_sort(Graph, AllVars).
build_graph([], Graph, Graph).
build_graph([V|AllVars0], Graph0, Graph) :-
clpbn:get_atts(V, [dist(_DistId, Parents)]), !,
dgraph_add_vertex(Graph0, V, Graph1),
add_parents(Parents, V, Graph1, GraphI),
build_graph(AllVars0, GraphI, Graph).
build_graph(_V.AllVars0, Graph0, Graph) :-
build_graph(AllVars0, Graph0, Graph).
add_parents([], _V, Graph, Graph).
add_parents([V0|Parents], V, Graph0, GraphF) :-
dgraph_add_edge(Graph0, V0, V, GraphI),
add_parents(Parents, V, GraphI, GraphF).
get_keys_info([], _, _, _, Vs, Vs, Ps, Ps, _, _) --> [].
get_keys_info([V|MoreVs], Evs, Fs, OrderVs, Vs, VsF, Ps, PsF, Lvs, Outs) -->
{ rb_lookup(V, F, Fs) }, !,
{ F = fn([V|Parents], _, _, DistId, _) },
%{writeln(v:DistId:Parents)},
[DIST],
{ get_key_info(V, F, Fs, Evs, OrderVs, DistId, Parents, Vs, Vs2, Ps, Ps1, Lvs, Outs, DIST) },
get_keys_info(MoreVs, Evs, Fs, OrderVs, Vs2, VsF, Ps1, PsF, Lvs, Outs).
get_key_info(V, F, Fs, Evs, OrderVs, DistId, Parents0, Vs, Vs2, Ps, Ps1, Lvs, Outs, DIST) :-
reorder_keys(Parents0, OrderVs, Parents, Map),
check_key_p(DistId, F, Map, Parms, _ParmVars, Ps, Ps1),
unbound_parms(Parms, ParmVars),
F = fn(_,[Size|_],_,_,_),
check_key(V, Size, DIST, Vs, Vs1),
DIST = info(V, Tree, Ev, Values, Formula, ParmVars, Parms),
% get a list of form [[P00,P01], [P10,P11], [P20,P21]]
foldl(get_key_parent(Fs), Parents, PVars, Vs1, Vs2),
cross_product(Values, Ev, PVars, ParmVars, Formula0),
% (numbervars(Formula0,0,_),writeln(formula0:Ev:Formula0), fail ; true),
get_key_evidence(V, Evs, DistId, Tree, Ev, Formula0, Formula, Lvs, Outs).
% (numbervars(Formula,0,_),writeln(formula:Formula), fail ; true).
get_vars_info([], Vs, Vs, Ps, Ps, _, _) --> [].
get_vars_info([V|MoreVs], Vs, VsF, Ps, PsF, Lvs, Outs) -->
{ clpbn:get_atts(V, [dist(DistId, Parents)]) }, !,
%{writeln(v:DistId:Parents)},
[DIST],
{ get_var_info(V, DistId, Parents, Vs, Vs2, Ps, Ps1, Lvs, Outs, DIST) },
get_vars_info(MoreVs, Vs2, VsF, Ps1, PsF, Lvs, Outs).
get_vars_info([_|MoreVs], Vs0, VsF, Ps0, PsF, VarsInfo, Lvs, Outs) :-
get_vars_info(MoreVs, Vs0, VsF, Ps0, PsF, VarsInfo, Lvs, Outs).
%
% let's have some fun with avg
%
get_var_info(V, avg(Domain), Parents, Vs, Vs2, Ps, Ps, Lvs, Outs, DIST) :- !,
length(Domain, DSize),
% run_though_avg(V, DSize, Domain, Parents, Vs, Vs2, Lvs, Outs, DIST).
% top_down_with_tabling(V, DSize, Domain, Parents, Vs, Vs2, Lvs, Outs, DIST).
bup_avg(V, DSize, Domain, Parents, Vs, Vs2, Lvs, Outs, DIST).
% standard random variable
get_var_info(V, DistId, Parents0, Vs, Vs2, Ps, Ps1, Lvs, Outs, DIST) :-
% clpbn:get_atts(V, [key(K)]), writeln(V:K:DistId:Parents),
reorder_vars(Parents0, Parents, Map),
check_p(DistId, Map, Parms, _ParmVars, Ps, Ps1),
unbound_parms(Parms, ParmVars),
check_v(V, DistId, DIST, Vs, Vs1),
DIST = info(V, Tree, Ev, Values, Formula, ParmVars, Parms),
% get a list of form [[P00,P01], [P10,P11], [P20,P21]]
get_parents(Parents, PVars, Vs1, Vs2),
cross_product(Values, Ev, PVars, ParmVars, Formula0),
% (numbervars(Formula0,0,_),writeln(formula0:Ev:Formula0), fail ; true),
get_evidence(V, Tree, Ev, Formula0, Formula, Lvs, Outs).
%, (numbervars(Formula,0,_),writeln(formula:Formula), fail ; true)
%
% reorder all variables and make sure we get a
% map of how the transfer was done.
%
% position zero is output
%
reorder_keys(Vs, Order, OVs, Map) :-
foldl(add_key_pos(Order), Vs, PVs, 1, _),
keysort(PVs, SVs),
maplist(remove_key,SVs, OVs, Map).
add_key_pos(Order, V, K-(I0,V), I0, I) :-
rb_lookup(V, K, Order),
I is I0+1.
reorder_vars(Vs, OVs, Map) :-
foldl(add_pos, Vs, PVs, 1, _),
keysort(PVs, SVs),
maplist(remove_key, SVs, OVs, Map).
add_pos(V, K-(I0,V), I0, I) :-
get_atts(V,[order(K)]),
I is I0+1.
remove_key(_-(I,V), V, I).
%%%%%%%%%%%%%%%%%%%%%%%%%
%
% use top-down to generate average
%
run_though_avg(V, 3, Domain, Parents0, Vs, Vs2, Lvs, Outs, DIST) :-
reorder_vars(Parents0, Parents, _Map),
check_v(V, avg(Domain,Parents0), DIST, Vs, Vs1),
DIST = info(V, Tree, Ev, [V0,V1,V2], Formula, [], []),
get_parents(Parents, PVars, Vs1, Vs2),
length(Parents, N),
generate_3tree(F00, PVars, 0, 0, 0, N, N0, N1, N2, R, (N1+2*N2 =< N/2), (N1+2*(N2+R) =< N/2)),
simplify_exp(F00, F0),
% generate_3tree(F1, PVars, 0, 0, 0, N, N0, N1, N2, R, ((N1+2*(N2+R) > N/2, N1+2*N2 < (3*N)/2))),
generate_3tree(F20, PVars, 0, 0, 0, N, N0, N1, N2, R, (N1+2*(N2+R) >= (3*N)/2), N1+2*N2 >= (3*N)/2),
% simplify_exp(F20, F2),
F20=F2,
Formula0 = [V0=F0*Ev0,V2=F2*Ev2,V1=not(F0+F2)*Ev1],
Ev = [Ev0,Ev1,Ev2],
get_evidence(V, Tree, Ev, Formula0, Formula, Lvs, Outs).
generate_3tree(OUT, _, I00, I10, I20, IR0, N0, N1, N2, R, _Exp, ExpF) :-
IR is IR0-1,
satisf(I00, I10, I20, IR, N0, N1, N2, R, ExpF),
!,
OUT = 1.
generate_3tree(OUT, [[P0,P1,P2]], I00, I10, I20, IR0, N0, N1, N2, R, Exp, _ExpF) :-
IR is IR0-1,
( satisf(I00+1, I10, I20, IR, N0, N1, N2, R, Exp) ->
L0 = [P0|L1]
;
L0 = L1
),
( satisf(I00, I10+1, I20, IR, N0, N1, N2, R, Exp) ->
L1 = [P1|L2]
;
L1 = L2
),
( satisf(I00, I10, I20+1, IR, N0, N1, N2, R, Exp) ->
L2 = [P2]
;
L2 = []
),
to_disj(L0, OUT).
generate_3tree(OUT, [[P0,P1,P2]|Ps], I00, I10, I20, IR0, N0, N1, N2, R, Exp, ExpF) :-
IR is IR0-1,
( satisf(I00+1, I10, I20, IR, N0, N1, N2, R, Exp) ->
I0 is I00+1, generate_3tree(O0, Ps, I0, I10, I20, IR, N0, N1, N2, R, Exp, ExpF)
->
L0 = [P0*O0|L1]
;
L0 = L1
),
( satisf(I00, I10+1, I20, IR0, N0, N1, N2, R, Exp) ->
I1 is I10+1, generate_3tree(O1, Ps, I00, I1, I20, IR, N0, N1, N2, R, Exp, ExpF)
->
L1 = [P1*O1|L2]
;
L1 = L2
),
( satisf(I00, I10, I20+1, IR0, N0, N1, N2, R, Exp) ->
I2 is I20+1, generate_3tree(O2, Ps, I00, I10, I2, IR, N0, N1, N2, R, Exp, ExpF)
->
L2 = [P2*O2]
;
L2 = []
),
to_disj(L0, OUT).
satisf(I0, I1, I2, IR, N0, N1, N2, R, Exp) :-
\+ \+ ( I0 = N0, I1=N1, I2=N2, IR=R, call(Exp) ).
not_satisf(I0, I1, I2, IR, N0, N1, N2, R, Exp) :-
\+ ( I0 = N0, I1=N1, I2=N2, IR=R, call(Exp) ).
%%%%%%%%%%%%%%%%%%%%%%%%%
%
% use top-down to generate average
%
top_down_with_tabling(V, Size, Domain, Parents0, Vs, Vs2, Lvs, Outs, DIST) :-
reorder_vars(Parents0, Parents, _Map),
check_v(V, avg(Domain,Parents), DIST, Vs, Vs1),
DIST = info(V, Tree, Ev, OVs, Formula, [], []),
get_parents(Parents, PVars, Vs1, Vs2),
length(Parents, N),
Max is (Size-1)*N, % This should be true
avg_borders(0, Size, Max, Borders),
b_hash_new(H0),
avg_trees(0, Max, PVars, Size, F1, 0, Borders, OVs, Ev, H0, H),
generate_avg_code(H, Formula, F),
% Formula0 = [V0=F0*Ev0,V2=F2*Ev2,V1=not(F0+F2)*Ev1],
% Ev = [Ev0,Ev1,Ev2],
get_evidence(V, Tree, Ev, F1, F, Lvs, Outs).
avg_trees(Size, _, _, Size, F0, _, F0, [], [], H, H) :- !.
avg_trees(I0, Max, PVars, Size, [V=O*E|F0], Im, [IM|Borders], [V|OVs], [E|Ev], H0, H) :-
I is I0+1,
avg_tree(PVars, 0, Max, Im, IM, Size, O, H0, HI),
Im1 is IM+1,
avg_trees(I, Max, PVars, Size, F0, Im1, Borders, OVs, Ev, HI, H).
avg_tree( _PVars, P, _, Im, IM, _Size, O, H0, H0) :-
b_hash_lookup(k(P,Im,IM), O=_Exp, H0), !.
avg_tree([], _P, _Max, _Im, _IM, _Size, 1, H, H).
avg_tree([Vals|PVars], P, Max, Im, IM, Size, O, H0, HF) :-
b_hash_insert(H0, k(P,Im,IM), O=Simp*1, HI),
MaxI is Max-(Size-1),
avg_exp(Vals, PVars, 0, P, MaxI, Size, Im, IM, HI, HF, Exp),
simplify_exp(Exp, Simp).
avg_exp([], _, _, _P, _Max, _Size, _Im, _IM, H, H, 0).
avg_exp([Val|Vals], PVars, I0, P0, Max, Size, Im, IM, HI, HF, O) :-
(Vals = [] -> O=O1 ; O = Val*O1+not(Val)*O2 ),
Im1 is max(0, Im-I0),
IM1 is IM-I0,
( IM1 < 0 -> O1 = 0, H2 = HI ; /* we have exceed maximum */
Im1 > Max -> O1 = 0, H2 = HI ; /* we cannot make to minimum */
Im1 = 0, IM1 > Max -> O1 = 1, H2 = HI ; /* we cannot exceed maximum */
P is P0+1,
avg_tree(PVars, P, Max, Im1, IM1, Size, O1, HI, H2)
),
I is I0+1,
avg_exp(Vals, PVars, I, P0, Max, Size, Im, IM, H2, HF, O2).
generate_avg_code(H, Formula, Formula0) :-
b_hash_to_list(H,L),
sort(L, S),
strip_and_add(S, Formula0, Formula).
strip_and_add([], F, F).
strip_and_add([_-Exp|S], F0, F) :-
strip_and_add(S, [Exp|F0], F).
%%%%%%%%%%%%%%%%%%%%%%%%%
%
% use bottom-up dynamic programming to generate average
%
bup_avg(V, Size, Domain, Parents0, Vs, Vs2, Lvs, Outs, DIST) :-
reorder_vars(Parents0, Parents, _),
check_v(V, avg(Domain,Parents), DIST, Vs, Vs1),
DIST = info(V, Tree, Ev, OVs, Formula, [], []),
get_parents(Parents, PVars, Vs1, Vs2),
length(Parents, N),
Max is (Size-1)*N, % This should be true
ArraySize is Max+1,
functor(Protected, protected, ArraySize),
avg_domains(0, Size, 0, Max, LDomains),
Domains =.. [d|LDomains],
Reach is (Size-1),
generate_sums(PVars, Size, Max, Reach, Protected, Domains, ArraySize, Sums, F0),
% bin_sums(PVars, Sums, F00),
% reverse(F00,F0),
% easier to do recursion on lists
Sums =.. [_|LSums],
generate_avg(0, Size, 0, Max, LSums, OVs, Ev, F1, []),
reverse(F0, RF0),
get_evidence(V, Tree, Ev, F1, F2, Lvs, Outs),
append(RF0, F2, Formula).
%
% use binary approach, like what is standard
%
bin_sums(Vs, Sums, F) :-
vs_to_sums(Vs, Sums0),
bin_sums(Sums0, Sums, F, []).
vs_to_sums([], []).
vs_to_sums([V|Vs], [Sum|Sums0]) :-
Sum =.. [sum|V],
vs_to_sums(Vs, Sums0).
bin_sums([Sum], Sum) --> !.
bin_sums(LSums, Sum) -->
{ halve(LSums, Sums1, Sums2) },
bin_sums(Sums1, Sum1),
bin_sums(Sums2, Sum2),
sum(Sum1, Sum2, Sum).
halve(LSums, Sums1, Sums2) :-
length(LSums, L),
Take is L div 2,
head(Take, LSums, Sums1, Sums2).
head(0, L, [], L) :- !.
head(Take, [H|L], [H|Sums1], Sum2) :-
Take1 is Take-1,
head(Take1, L, Sums1, Sum2).
sum(Sum1, Sum2, Sum) -->
{ functor(Sum1, _, M1),
functor(Sum2, _, M2),
Max is M1+M2-2,
Max1 is Max+1,
Max0 is M2-1,
functor(Sum, sum, Max1),
Sum1 =.. [_|PVals] },
expand_sums(PVals, 0, Max0, Max1, M2, Sum2, Sum).
%
% bottom up step by step
%
%
generate_sums([PVals], Size, Max, _, _Protected, _Domains, _, Sum, []) :- !,
Max is Size-1,
Sum =.. [sum|PVals].
generate_sums([PVals|Parents], Size, Max, Reach, Protected, Domains, ASize, NewSums, F) :-
NewReach is Reach+(Size-1),
generate_sums(Parents, Size, Max0, NewReach, Protected, Domains, ASize, Sums, F0),
Max is Max0+(Size-1),
Max1 is Max+1,
functor(NewSums, sum, Max1),
protect_avg(0, Max0, Protected, Domains, ASize, Reach),
expand_sums(PVals, 0, Max0, Max1, Size, Sums, Protected, NewSums, F, F0).
protect_avg(Max0,Max0,_Protected, _Domains, _ASize, _Reach) :- !.
protect_avg(I0, Max0, Protected, Domains, ASize, Reach) :-
I is I0+1,
Top is I+Reach,
( Top > ASize ;
arg(I, Domains, CD),
arg(Top, Domains, CD)
), !,
arg(I, Protected, yes),
protect_avg(I, Max0, Protected, Domains, ASize, Reach).
protect_avg(I0, Max0, Protected, Domains, ASize, Reach) :-
I is I0+1,
protect_avg(I, Max0, Protected, Domains, ASize, Reach).
%
% outer loop: generate array of sums at level j= Sum[j0...jMax]
%
expand_sums(_Parents, Max, _, Max, _Size, _Sums, _P, _NewSums, F0, F0) :- !.
expand_sums(Parents, I0, Max0, Max, Size, Sums, Prot, NewSums, [O=SUM*1|F], F0) :-
I is I0+1,
arg(I, Prot, P),
var(P), !,
arg(I, NewSums, O),
sum_all(Parents, 0, I0, Max0, Sums, List),
to_disj(List, SUM),
expand_sums(Parents, I, Max0, Max, Size, Sums, Prot, NewSums, F, F0).
expand_sums(Parents, I0, Max0, Max, Size, Sums, Prot, NewSums, F, F0) :-
I is I0+1,
arg(I, Sums, O),
arg(I, NewSums, O),
expand_sums(Parents, I, Max0, Max, Size, Sums, Prot, NewSums, F, F0).
%
%inner loop: find all parents that contribute to A_ji,
% that is generate Pk*Sum_(j-1)l and k+l st k+l = i
%
sum_all([], _, _, _, _, []).
sum_all([V|Vs], Pos, I, Max0, Sums, [O|List]) :-
J is I-Pos,
J >= 0,
J =< Max0, !,
J1 is J+1,
arg(J1, Sums, S0),
( J < I -> O = V*S0 ; O = S0*V ),
Pos1 is Pos+1,
sum_all(Vs, Pos1, I, Max0, Sums, List).
sum_all([_V|Vs], Pos, I, Max0, Sums, List) :-
Pos1 is Pos+1,
sum_all(Vs, Pos1, I, Max0, Sums, List).
gen_arg(J, Sums, Max, S0) :-
gen_arg(0, Max, J, Sums, S0).
gen_arg(Max, Max, J, Sums, S0) :- !,
I is Max+1,
arg(I, Sums, A),
( Max = J -> S0 = A ; S0 = not(A)).
gen_arg(I0, Max, J, Sums, S) :-
I is I0+1,
arg(I, Sums, A),
( I0 = J -> S = A*S0 ; S = not(A)*S0),
gen_arg(I, Max, J, Sums, S0).
avg_borders(Size, Size, _Max, []) :- !.
avg_borders(I0, Size, Max, [J|Vals]) :-
I is I0+1,
Border is (I*Max)/Size,
J is integer(round(Border)),
avg_borders(I, Size, Max, Vals).
avg_domains(Size, Size, _J, _Max, []).
avg_domains(I0, Size, J0, Max, Vals) :-
I is I0+1,
Border is (I*Max)/Size,
fetch_domain_for_avg(J0, Border, J, I0, Vals, ValsI),
avg_domains(I, Size, J, Max, ValsI).
fetch_domain_for_avg(J, Border, J, _, Vals, Vals) :-
J > Border, !.
fetch_domain_for_avg(J0, Border, J, I0, [I0|LVals], RLVals) :-
J1 is J0+1,
fetch_domain_for_avg(J1, Border, J, I0, LVals, RLVals).
generate_avg(Size, Size, _J, _Max, [], [], [], F, F).
generate_avg(I0, Size, J0, Max, LSums, [O|OVs], [Ev|Evs], [O=Disj*Ev|F], F0) :-
I is I0+1,
Border is (I*Max)/Size,
fetch_for_avg(J0, Border, J, LSums, MySums, RSums),
to_disj(MySums, Disj),
generate_avg(I, Size, J, Max, RSums, OVs, Evs, F, F0).
fetch_for_avg(J, Border, J, RSums, [], RSums) :-
J > Border, !.
fetch_for_avg(J0, Border, J, [S|LSums], [S|MySums], RSums) :-
J1 is J0+1,
fetch_for_avg(J1, Border, J, LSums, MySums, RSums).
to_disj([], 0).
to_disj([V], V).
to_disj([V,V1|Vs], Out) :-
to_disj2([V1|Vs], V, Out).
to_disj2([V], V0, V0+V).
to_disj2([V,V1|Vs], V0, Out) :-
to_disj2([V1|Vs], V0+V, Out).
%
% look for parameters in the rb-tree, or add a new.
% distid is the key
%
check_key_p(DistId, _, Map, Parms, ParmVars, Ps, Ps) :-
rb_lookup(DistId-Map, theta(Parms, ParmVars), Ps), !.
check_key_p(DistId, fn(_, Sizes, Parms0, DistId, _), Map, Parms, ParmVars, Ps, PsF) :-
swap_parms(Parms0, Sizes, [0|Map], Parms1),
length(Parms1, L0),
Sizes = [Size|_],
L1 is L0 div Size,
L is L0-L1,
initial_maxes(L1, Multipliers),
copy(L, Multipliers, NextMults, NextMults, Parms1, Parms, ParmVars),
%writeln(t:Size:Parms0:Parms:ParmVars),
rb_insert(Ps, DistId-Map, theta(Parms, ParmVars), PsF).
%
% look for parameters in the rb-tree, or add a new.
% distid is the key
%
check_p(DistId, Map, Parms, ParmVars, Ps, Ps) :-
rb_lookup(DistId-Map, theta(Parms, ParmVars), Ps), !.
check_p(DistId, Map, Parms, ParmVars, Ps, PsF) :-
get_dist_params(DistId, Parms0),
get_dist_all_sizes(DistId, Sizes),
swap_parms(Parms0, Sizes, [0|Map], Parms1),
length(Parms1, L0),
get_dist_domain_size(DistId, Size),
L1 is L0 div Size,
L is L0-L1,
initial_maxes(L1, Multipliers),
copy(L, Multipliers, NextMults, NextMults, Parms1, Parms, ParmVars),
%writeln(t:Size:Parms0:Parms:ParmVars),
rb_insert(Ps, DistId-Map, theta(Parms, ParmVars), PsF).
swap_parms(Parms0, Sizes, Map, Parms1) :-
matrix_new(floats, Sizes, Parms0, T0),
matrix_shuffle(T0,Map,TF),
matrix_to_list(TF, Parms1).
%
% we are using switches by two
%
initial_maxes(0, []) :- !.
initial_maxes(Size, [1.0|Multipliers]) :- !,
Size1 is Size-1,
initial_maxes(Size1, Multipliers).
copy(0, [], [], _, _Parms0, [], []) :- !.
copy(N, [], [], Ms, Parms0, Parms, ParmVars) :-!,
copy(N, Ms, NewMs, NewMs, Parms0, Parms, ParmVars).
copy(N, D.Ds, ND.NDs, New, El.Parms0, NEl.Parms, V.ParmVars) :-
N1 is N-1,
(El == 0.0 ->
NEl = 0,
V = NEl,
ND = D
;El == 1.0 ->
NEl = 1,
V = NEl,
ND = 0.0
;El == 0 ->
NEl = 0,
V = NEl,
ND = D
;El =:= 1 ->
NEl = 1,
V = NEl,
ND = 0.0,
V = NEl
;
NEl is El/D,
ND is D-El,
V = NEl
),
copy(N1, Ds, NDs, New, Parms0, Parms, ParmVars).
unbound_parms([], []).
unbound_parms(_.Parms, _.ParmVars) :-
unbound_parms(Parms, ParmVars).
check_v(V, _, INFO, Vs, Vs) :-
rb_lookup(V, INFO, Vs), !.
check_v(V, DistId, INFO, Vs0, Vs) :-
get_dist_domain_size(DistId, Size),
length(Values, Size),
length(Ev, Size),
INFO = info(V, _Tree, Ev, Values, _Formula, _, _),
rb_insert(Vs0, V, INFO, Vs).
get_parents([], [], Vs, Vs).
get_parents(V.Parents, Values.PVars, Vs0, Vs) :-
clpbn:get_atts(V, [dist(DistId, _)]),
check_v(V, DistId, INFO, Vs0, Vs1),
INFO = info(V, _Parent, _Ev, Values, _, _, _),
get_parents(Parents, PVars, Vs1, Vs).
get_key_parent(Fs, V, Values, Vs0, Vs) :-
INFO = info(V, _Parent, _Ev, Values, _, _, _),
rb_lookup(V, fn(_, [Size|_], _, _, _), Fs),
check_key(V, Size, INFO, Vs0, Vs).
check_key(V, _, INFO, Vs, Vs) :-
rb_lookup(V, INFO, Vs), !.
check_key(V, Size, INFO, Vs0, Vs) :-
length(Values, Size),
length(Ev, Size),
INFO = info(V, _Tree, Ev, Values, _Formula, _, _),
rb_insert(Vs0, V, INFO, Vs).
%
% construct the formula, this is the key...
%
cross_product(Values, Ev, PVars, ParmVars, Formulas) :-
arrangements(PVars, Arranges),
apply_parents_first(Values, Ev, ParmCombos, ParmCombos, Arranges, Formulas, ParmVars).
%
% if we have the parent variables with two values, we get
% [[XP,YP],[XP,YN],[XN,YP],[XN,YN]]
%
arrangements([], [[]]).
arrangements([L1|Ls],O) :-
arrangements(Ls, LN),
expand(L1, LN, O, []).
expand([], _LN) --> [].
expand([H|L1], LN) -->
concatenate_all(H, LN),
expand(L1, LN).
concatenate_all(_H, []) --> [].
concatenate_all(H, [L|LN]) -->
[[H|L]],
concatenate_all(H, LN).
%
% core of algorithm
%
% Values -> Output Vars for BDD
% Es -> Evidence variables
% Previous -> top of difference list with parameters used so far
% P0 -> end of difference list with parameters used so far
% Pvars -> Parents
% Eqs -> Output Equations
% Pars -> Output Theta Parameters
%
apply_parents_first([Value], [E], Previous, [], PVars, [Value=Disj*E], Parameters) :- !,
apply_last_parent(PVars, Previous, Disj),
flatten(Previous, Parameters).
apply_parents_first([Value|Values], [E|Ev], Previous, P0, PVars, (Value=Disj*E).Formulas, Parameters) :-
P0 = [TheseParents|End],
apply_first_parent(PVars, Disj, TheseParents),
apply_parents_second(Values, Ev, Previous, End, PVars, Formulas, Parameters).
apply_parents_second([Value], [E], Previous, [], PVars, [Value=Disj*E], Parameters) :- !,
apply_last_parent(PVars, Previous, Disj),
flatten(Previous, Parameters).
apply_parents_second([Value|Values], [E|Ev], Previous, P0, PVars, (Value=Disj*E).Formulas, Parameters) :-
apply_middle_parent(PVars, Previous, Disj, TheseParents),
% this must be done after applying middle parents because of the var
% test.
P0 = [TheseParents|End],
apply_parents_second(Values, Ev, Previous, End, PVars, Formulas, Parameters).
apply_first_parent([Parents], Conj, [Theta]) :- !,
parents_to_conj(Parents,Theta,Conj).
apply_first_parent(Parents.PVars, Conj+Disj, Theta.TheseParents) :-
parents_to_conj(Parents,Theta,Conj),
apply_first_parent(PVars, Disj, TheseParents).
apply_middle_parent([Parents], Other, Conj, [ThetaPar]) :- !,
skim_for_theta(Other, Theta, _, ThetaPar),
parents_to_conj(Parents,Theta,Conj).
apply_middle_parent(Parents.PVars, Other, Conj+Disj, ThetaPar.TheseParents) :-
skim_for_theta(Other, Theta, Remaining, ThetaPar),
parents_to_conj(Parents,(Theta),Conj),
apply_middle_parent(PVars, Remaining, Disj, TheseParents).
apply_last_parent([Parents], Other, Conj) :- !,
parents_to_conj(Parents,(Theta),Conj),
skim_for_theta(Other, Theta, _, _).
apply_last_parent(Parents.PVars, Other, Conj+Disj) :-
parents_to_conj(Parents,(Theta),Conj),
skim_for_theta(Other, Theta, Remaining, _),
apply_last_parent(PVars, Remaining, Disj).
%
%
% simplify stuff, removing process that is cancelled by 0s
%
parents_to_conj([], Theta, Theta) :- !.
parents_to_conj(Ps, Theta, Theta*Conj) :-
parents_to_conj2(Ps, Conj).
parents_to_conj2([P],P) :- !.
parents_to_conj2([P|Ps],P*Conj) :-
parents_to_conj2(Ps,Conj).
%
% first case we haven't reached the end of the list so we need
% to create a new parameter variable
%
skim_for_theta([[P|Other]|V], not(P)*New, [Other|_], New) :- var(V), !.
%
% last theta, it is just negation of the other ones
%
skim_for_theta([[P|Other]], not(P), [Other], _) :- !.
%
% recursive case, build-up
%
skim_for_theta([[P|Other]|More], not(P)*Ps, [Other|Left], New ) :-
skim_for_theta(More, Ps, Left, New ).
get_key_evidence(V, Evs, _, Tree, Ev, F0, F, Leaves, Finals) :-
rb_lookup(V, Pos, Evs), !,
zero_pos(0, Pos, Ev),
insert_output(Leaves, V, Finals, Tree, Outs, SendOut),
get_outs(F0, F, SendOut, Outs).
% hidden deterministic node, can be removed.
%% get_key_evidence(V, _, DistId, _Tree, Ev, F0, [], _Leaves, _Finals) :-
%% deterministic(V, DistId),
%% !,
%% one_list(Ev),
%% eval_outs(F0).
%% no evidence !!!
get_key_evidence(V, _, _, Tree, _Values, F0, F1, Leaves, Finals) :-
insert_output(Leaves, V, Finals, Tree, Outs, SendOut),
get_outs(F0, F1, SendOut, Outs).
get_evidence(V, Tree, Ev, F0, F, Leaves, Finals) :-
clpbn:get_atts(V, [evidence(Pos)]), !,
zero_pos(0, Pos, Ev),
insert_output(Leaves, V, Finals, Tree, Outs, SendOut),
get_outs(F0, F, SendOut, Outs).
% hidden deterministic node, can be removed.
get_evidence(V, _Tree, Ev, F0, [], _Leaves, _Finals) :-
clpbn:get_atts(V, [key(K)]),
functor(K, Name, 2),
( Name = 'AVG' ; Name = 'MAX' ; Name = 'MIN' ),
!,
one_list(Ev),
eval_outs(F0).
%% no evidence !!!
get_evidence(V, Tree, _Values, F0, F1, Leaves, Finals) :-
insert_output(Leaves, V, Finals, Tree, Outs, SendOut),
get_outs(F0, F1, SendOut, Outs).
zero_pos(_, _Pos, []).
zero_pos(Pos, Pos, [1|Values]) :- !,
I is Pos+1,
zero_pos(I, Pos, Values).
zero_pos(I0, Pos, [0|Values]) :-
I is I0+1,
zero_pos(I, Pos, Values).
one_list([]).
one_list(1.Ev) :-
one_list(Ev).
%
% insert a node with the disj of all alternatives, this is only done if node ends up to be in the output
%
insert_output([], _V, [], _Out, _Outs, []).
insert_output(V._Leaves, V0, [Top|_], Top, Outs, [Top = Outs]) :- V == V0, !.
insert_output(_.Leaves, V, _.Finals, Top, Outs, SendOut) :-
insert_output(Leaves, V, Finals, Top, Outs, SendOut).
get_outs([V=F], [V=NF|End], End, V) :- !,
% writeln(f0:F),
simplify_exp(F,NF).
get_outs([(V=F)|Outs], [(V=NF)|NOuts], End, (F0 + V)) :-
% writeln(f0:F),
simplify_exp(F,NF),
get_outs(Outs, NOuts, End, F0).
eval_outs([]).
eval_outs([(V=F)|Outs]) :-
simplify_exp(F,NF),
V = NF,
eval_outs(Outs).
run_solver(Qs, LLPs, bdd(Term, Leaves, Nodes, Hash, Id)) :-
lists_of_keys_to_ids(Qs, QIds, Hash, _, Id, _),
findall(LPs,
(member(Q, QIds),
run_bdd_solver([Q],LPs,bdd(Term,Leaves,Nodes))),
LLPs).
run_bdd_solver([Vs], LPs, bdd(Term, _Leaves, Nodes)) :-
build_out_node(Nodes, Node),
findall(Prob, get_prob(Term, Node, Vs, Prob),TermProbs),
sumlist(TermProbs, Sum),
normalise(TermProbs, Sum, LPs).
% output node for BDDs
build_out_node([_Top], []).
build_out_node([T,T1|Tops], [_Top = T*NTop]) :-
build_out_node2([T1|Tops], NTop).
build_out_node2([Top], Top).
build_out_node2([T,T1|Tops], T*Top) :-
build_out_node2(T1.Tops, Top).
get_prob(Term, _Node, Vs, SP) :-
bdds(ddnnf), !,
all_cnfs(Term, CNF, IVs, Indics, Vs, AllParms, AllParmValues),
build_cnf(CNF, IVs, Indics, AllParms, AllParmValues, SP).
get_prob(Term, Node, Vs, SP) :-
bdds(bdd), !,
bind_all(Term, Node, Bindings, Vs, AllParms, AllParmValues),
% reverse(AllParms, RAllParms),
term_variables(AllParms, NVs),
build_bdd(Bindings, NVs, AllParms, AllParmValues, Bdd),
bdd_to_probability_sum_product(Bdd, SP),
bdd_close(Bdd).
build_bdd(Bindings, NVs, VTheta, Theta, Bdd) :-
bdd_from_list(Bindings, NVs, Bdd),
% bdd_size(Bdd, Len),
% number_codes(Len,Codes),
% atom_codes(Name,Codes),
% bdd_print(Bdd, Name),
% writeln(length=Len),
VTheta = Theta.
bind_all([], End, End, _V, [], []).
bind_all([info(V, _Tree, Ev, _Values, Formula, ParmVars, Parms)|Term], End, BindsF, V0s, ParmVars.AllParms, Parms.AllTheta) :-
v_in(V, V0s), !,
set_to_one_zeros(Ev),
bind_formula(Formula, BindsF, BindsI),
bind_all(Term, End, BindsI, V0s, AllParms, AllTheta).
bind_all([info(_V, _Tree, Ev, _Values, Formula, ParmVars, Parms)|Term], End, BindsF, V0s, ParmVars.AllParms, Parms.AllTheta) :-
set_to_ones(Ev),!,
bind_formula(Formula, BindsF, BindsI),
bind_all(Term, End, BindsI, V0s, AllParms, AllTheta).
% evidence: no need to add any stuff.
bind_all([info(_V, _Tree, _Ev, _Values, Formula, ParmVars, Parms)|Term], End, BindsF, V0s, ParmVars.AllParms, Parms.AllTheta) :-
bind_formula(Formula, BindsF, BindsI),
bind_all(Term, End, BindsI, V0s, AllParms, AllTheta).
bind_formula([], L, L).
bind_formula([B|Formula], [B|BsF], Bs0) :-
bind_formula(Formula, BsF, Bs0).
set_to_one_zeros([1|Values]) :-
set_to_zeros(Values).
set_to_one_zeros([0|Values]) :-
set_to_one_zeros(Values).
set_to_zeros([]).
set_to_zeros([0|Values]) :-
set_to_zeros(Values).
set_to_ones([]).
set_to_ones([1|Values]) :-
set_to_ones(Values).
normalise([], _Sum, []).
normalise([P|TermProbs], Sum, [NP|LPs]) :-
NP is P/Sum,
normalise(TermProbs, Sum, LPs).
finalize_bdd_solver(_).
all_cnfs([], [], [], [], _V, [], []).
all_cnfs([info(V, Tree, Ev, Values, Formula, ParmVars, Parms)|Term], BindsF, IVars, Indics, V0s, AllParmsF, AllThetaF) :-
%writeln(f:Formula),
v_in(V, V0s), !,
set_to_one_zeros(Ev),
all_indicators(Values, BindsF, Binds0),
indicators(Values, [], Ev, IVars, IVarsI, Indics, IndicsI, Binds0, Binds1),
parms( ParmVars, Parms, AllParmsF, AllThetaF, AllParms, AllTheta),
parameters(Formula, Tree, Binds1, BindsI),
all_cnfs(Term, BindsI, IVarsI, IndicsI, V0s, AllParms, AllTheta).
all_cnfs([info(_V, Tree, Ev, Values, Formula, ParmVars, Parms)|Term], BindsF, IVars, Indics, V0s, AllParmsF, AllThetaF) :-
set_to_ones(Ev),!,
all_indicators(Values, BindsF, Binds0),
indicators(Values, [], Ev, IVars, IVarsI, Indics, IndicsI, Binds0, Binds1),
parms( ParmVars, Parms, AllParmsF, AllThetaF, AllParms, AllTheta),
parameters(Formula, Tree, Binds1, BindsI),
all_cnfs(Term, BindsI, IVarsI, IndicsI, V0s, AllParms, AllTheta).
% evidence: no need to add any stuff.
all_cnfs([info(_V, Tree, Ev, Values, Formula, ParmVars, Parms)|Term], BindsF, IVars, Indics, V0s, AllParmsF, AllThetaF) :-
all_indicators(Values, BindsF, Binds0),
indicators(Values, [], Ev, IVars, IVarsI, Indics, IndicsI, Binds0, Binds1),
parms( ParmVars, Parms, AllParmsF, AllThetaF, AllParms, AllTheta),
parameters(Formula, Tree, Binds1, BindsI),
all_cnfs(Term, BindsI, IVarsI, IndicsI, V0s, AllParms, AllTheta).
v_in(V, [V0|_]) :- V == V0, !.
v_in(V, [_|Vs]) :-
v_in(V, Vs).
all_indicators(Values) -->
{ values_to_disj(Values, Disj) },
[Disj].
values_to_disj([V], V) :- !.
values_to_disj([V|Values], V+Disj) :-
values_to_disj(Values, Disj).
indicators([V|Vars], SeenVs, [E|Ev], [V|IsF], IsI, [E|Inds], Inds0) -->
generate_exclusions(SeenVs, V),
indicators(Vars, [V|SeenVs], Ev, IsF, IsI, Inds, Inds0).
indicators([], _SeenVs, [], IsF, IsF, Inds, Inds) --> [].
parms([], [], AllParms, AllTheta, AllParms, AllTheta).
parms([V|ParmVars], [P|Parms], [V|AllParmsF], [P|AllThetaF], AllParms, AllTheta) :-
parms( ParmVars, Parms, AllParmsF, AllThetaF, AllParms, AllTheta).
parameters([], _) --> [].
% ignore disj, only useful to BDDs
parameters([(T=_)|Formula], Tree) -->
{ Tree == T }, !,
parameters(Formula, Tree).
parameters([(V0=Disj*_I0)|Formula], Tree) -->
conj(Disj, V0),
parameters(Formula, Tree).
% transform V0<- A*B+C*(D+not(E))
% [V0+not(A)+not(B),V0+not(C)+not(D),V0+not(C)+E]
conj(Disj, V0) -->
{ conj2(Disj, [[V0]], LVs) },
to_disjs(LVs).
conj2(A, L0, LF) :- var(A), !,
add(not(A), L0, LF).
conj2((A*B), L0, LF) :-
conj2(A, L0, LI),
conj2(B, LI, LF).
conj2((A+B), L0, LF) :-
conj2(A, L0, L1),
conj2(B, L0, L2),
append(L1, L2, LF).
conj2(not(A), L0, LF) :-
add(A, L0, LF).
add(_, [], []).
add(Head, [H|L], [[Head|H]|NL]) :-
add(Head, L, NL).
to_disjs([]) --> [].
to_disjs([[H|L]|LVs]) -->
mkdisj(L, H),
to_disjs(LVs).
mkdisj([], Disj) --> [Disj].
mkdisj([H|L], Disj) -->
mkdisj(L, (H+Disj)).
%
% add formula for V \== V0 -> V or V0 and not(V) or not(V0)
%
generate_exclusions([], _V) --> [].
generate_exclusions([V0|SeenVs], V) -->
[(not(V0)+not(V))],
generate_exclusions(SeenVs, V).
build_cnf(CNF, IVs, Indics, AllParms, AllParmValues, Val) :-
%(numbervars(CNF,1,_), writeln(cnf_to_ddnnf(CNF, Vars, IVs, [], F)), fail ; true ),
cnf_to_ddnnf(CNF, AllParms, F),
AllParms = AllParmValues,
IVs = Indics,
term_variables(CNF, Extra),
set_to_ones(Extra),
ddnnf_is(F, Val).