%% add BNT (may require changes)
current_dir=pwd;				%saves current working directory
cd 'C:\MATLAB\work\bnt-master';	%this should point to the BNT directory
addpath(genpathKPM(pwd));		%add BNT paths
cd(current_dir);				%return to working directory
%% create a synthetic dataset from Fig. 1 of Kelner and Lerner (2012)
clear all
clc
num_of_nodes=5;                 %total number of nodes
X1=1;                           %node names. Note: DAG must possess topological order.
X2=2;
C=3;
X3=4;
X4=5;
dag=zeros(num_of_nodes,num_of_nodes);   %five nodes 'X1','X2','C','X3','X4'
dag(X1,X2)=1;                           %an edge from X1 to X2
dag(X2,C)=1;                            %an edge from X2 to C
dag(X3,X4)=1;                           %an edge from X3 to X4
dag(C,X4)=1;                            %an edge from C to X4
discrete_nodes = 1:num_of_nodes;        %all nodes are discrete
node_sizes = 2*ones(1,num_of_nodes);    %all nodes have two possible values
%create BNT
bnet = mk_bnet(dag, node_sizes, 'names',{'X1','X2','C','X3','X4'},'discrete', discrete_nodes);
bnet.CPD{X1} = tabular_CPD(bnet, X1, [0.3 0.7]);
bnet.CPD{X3} = tabular_CPD(bnet, X3, [0.7 0.3]);
bnet.CPD{X2} = tabular_CPD(bnet, X2, [0.7 0.3 0.3 0.7]);
bnet.CPD{C} = tabular_CPD(bnet, C, [0.85 0.25 0.15 0.75]);
bnet.CPD{X4} = tabular_CPD(bnet, X4, [0.1 0.8 0.3 0.2 0.9 0.2 0.7 0.8]);
num_of_instances=1000;                   %set number of instances
my_train_samples=cell(num_of_nodes,num_of_instances);
my_test_samples=cell(num_of_nodes,num_of_instances);
%generate train and test datasets
for c1=1:num_of_instances
    my_train_samples(:,c1)=sample_bnet(bnet);
end
for c1=1:num_of_instances
    my_test_samples(:,c1)=sample_bnet(bnet);
end

%% learning an RMCV classifier
%initial DAG
init_dag=dag;
init_dag(C,X4)=0;

% %common alternatives:
% %empty DAG
% init_dag=zeros(num_of_nodes,num_of_nodes);
% %NBC
% init_dag=zeros(num_of_nodes,num_of_nodes);
% init_dag(C,[X1 X2 X3 X4])=1;

%learn BN using RMCV
[final_dag,history_dags,history_scores]=rmcv_gs(init_dag,my_train_samples,my_test_samples,node_sizes,C,4);