support for randomising or making uniform all distributions before starting EM

CLPBN/clpbn/dists.yap

@@ -20,8 +20,11 @@
 	   get_evidence_from_position/3,
 	   dist_to_term/2,
 	   empty_dist/2,
-	   dist_new_table/2,
+	   all_dist_ids/1,
-	   all_dist_ids/1
+	   randomise_all_dists/0,
+	   randomise_dist/1,
+	   uniformise_all_dists/0,
+	   uniformise_dist/1
 	]).
 
 :- use_module(library(lists),[is_list/1,nth0/3]).
@@ -32,6 +35,10 @@
 	       matrix_to_list/2,
 	       matrix_to_logs/1]).
 
+:- use_module(library('clpbn/matrix_cpt_utils'),
+	      [random_CPT/2,
+	       uniform_CPT/2]).
+
 /*
 :- mode dist(+, -).
 
@@ -258,3 +265,24 @@ copy_Lstructure([H|LKey], [NH|LKey0]) :-
 	copy_structure(H, NH),
 	copy_Lstructure(LKey, LKey0).
 
+randomise_all_dists :-
+	randomise_dist(_),
+	fail.
+randomise_all_dists.
+
+randomise_dist(Dist) :-
+	recorded(clpbn_dist_psizes, db(Dist,DSizes), _),
+	random_CPT(DSizes, NewCPT),
+	dist_new_table(Dist, NewCPT).
+
+uniformise_all_dists :-
+	uniformise_dist(_),
+	fail.
+uniformise_all_dists.
+
+uniformise_dist(Dist) :-
+	recorded(clpbn_dist_psizes, db(Dist,DSizes), _),
+	uniform_CPT(DSizes, NewCPT),
+	dist_new_table(Dist, NewCPT).
+
+

CLPBN/clpbn/matrix_cpt_utils.yap

@@ -15,7 +15,9 @@
 	   multiply_factors/3,
 	   normalise_possibly_deterministic_CPT/2,
 	   column_from_possibly_deterministic_CPT/3,
-	   multiply_possibly_deterministic_factors/3]).
+	   multiply_possibly_deterministic_factors/3,
+	   random_CPT/2,
+	   uniform_CPT/2]).
 
 :- use_module(dists,
 	      [get_dist_domain_size/2,
@@ -26,6 +28,7 @@
 	       matrix_new_set/4,
 	       matrix_select/4,
 	       matrix_dims/2,
+	       matrix_size/2,
 	       matrix_shuffle/3,
 	       matrix_expand/3,
 	       matrix_op/4,
@@ -229,3 +232,25 @@ normalise_possibly_deterministic_CPT(MAT,NMAT) :-
 	matrix_agg_lines(MAT, +, Sum),
 	matrix_op_to_lines(MAT, Sum, /, NMAT).
 
+random_CPT(Dims, M) :-
+	mult_all(Dims,1,Size),
+	generate_random_entries(Size, Randoms),
+	matrix_new(floats, Dims, Randoms, M1),
+	normalise_possibly_deterministic_CPT(M1, M).
+
+mult_all([],Size,Size).
+mult_all([D|Dims],Size0,Size) :-
+	Size1 is Size0*D,
+	mult_all(Dims,Size1,Size).
+
+generate_random_entries(0, []) :- !.
+generate_random_entries(Size, [R|Randoms]) :-
+	R is random,
+	Size1 is Size-1,
+	generate_random_entries(Size1, Randoms).
+
+uniform_CPT(Dims, M) :-
+	matrix_new_set(floats,Dims,1.0,M1),
+	normalise_possibly_deterministic_CPT(M1, M).
+
+	

CLPBN/learning/em.yap

@@ -16,7 +16,9 @@
 	      [get_dist_domain_size/2,
 	       empty_dist/2,
 	       dist_new_table/2,
-	       get_dist_key/2]).
+	       get_dist_key/2,
+	       randomise_all_dists/0,
+	       uniformise_all_dists/0]).
 
 :- use_module(library('clpbn/connected'),
 	      [clpbn_subgraphs/2]).
@@ -62,6 +64,8 @@ em(Items, MaxError, MaxIts, Tables, Likelihood) :-
 % and more detailed info on distributions, namely with a list of all instances for the distribution.
 init_em(Items, state( AllDists, AllDistInstances, MargVars, SolverVars)) :-
 	run_all(Items),
+%	randomise_all_dists,
+	uniformise_all_dists,
 	attributes:all_attvars(AllVars0),
 	sort_vars_by_key(AllVars0,AllVars1,[]),
 	% remove variables that do not have to do with this query.