inference.m

%% -------------------------------------------------------------------------
% ( 9 ) Inference & missing values
% -------------------------------------------------------------------------

% Create inference engine
engine = jtree_inf_engine ( bnet );

% Sort brokenData to right order
brokenDataSorted = brokenData ( bnet.ordering, : )';

% Find nanvalues in broken data
[nanrow, nancol] = find(isnan(brokenDataSorted));

% First we need to tranform brokenData into cell array
cbroken = num2cell ( brokenDataSorted );

[broken_num, ~] = size ( cbroken );

% Delete the NaN values from cell array
[cbroken{nanrow + broken_num .* (nancol-1)}] = deal([]);

% For all broken samples
for i = 1:broken_num
	% Get one sample
	csample = cbroken ( i, : );

	% Create evidence from sample
	evidence = csample;

	% Print evidence
	fprintf('\nAll evidence for %d-nth sample:\n', i);
	for e = 1:N
	    cEvid = evidence { e };
	    if (~isnan(cEvid))
	        fprintf ( '%15s = %-9s\n', data.names{bnet.ordering(e)}, getLabel( bnet.ordering(e), cEvid ));
	    end
	end
	fprintf ( '\n' );

	% Add evidence to the engine
	engine = enter_evidence ( engine, evidence );

	% All nan variables
	varIdxs = nancol ( nanrow == i );

	% Count all nan variables
	countNanVar = size ( varIdxs, 1 );

	% For all nan variables
	for j = 1:countNanVar

		% Find index of current nan variable
		varIdx = varIdxs ( j );

		% Result - Marginal distribution
		marg = marginal_nodes ( engine, varIdx );

		fprintf('\n');
		% Print all result
		for v = 1:size(marg.T)
			fprintf('Pr (%s = %-9s | all_others) = %f\n', data.names{bnet.ordering(varIdx)}, getLabel ( bnet.ordering(varIdx), v ), marg.T(v) );
		end

		% Maximum
		[mmax, midx] = max ( marg.T );	
		fprintf('\n');
		fprintf( 'The value of "%s" most probably is: %-9s\n', data.names{bnet.ordering(varIdx)}, getLabel ( bnet.ordering(varIdx), midx ) );
		
	end

end