function [sim] = simReBinary(probMatrix, AintersectB, AnotB, BnotA)
% Residual entropy method 
% (ref: Information-theoretic and Set-theoretic Similarity - Cazzanti and
% Gupta)

% Inputs: 
% probMatrix(m,n): m = number of possible values of any feature (2 for
%                       discrete case)
%                  n = number of features
% probMatrix(i,j) = probability of seeing value i in feature j
% AintersectB: intersection of feature vectors of a and b
% AnotB: features of a not in b
% BnotA: features of b not in a
% number of columns in probMatrix = length of AintersectB = length of AnotB
% = length BnotA

% Output: similarity value as calculated by residual entropy method

numValues = size(probMatrix,1);
numFeatures = size(probMatrix, 2);
sim = 0;
for i = 1:numValues
    intersect = 1;
    flagIntersect = 0;
    exclusiveA = 1;
    flagA = 0;
    exclusiveB = 1;
    flagB = 0;
    for j = 1:numFeatures
        if(AintersectB(j)~= 0 && flagIntersect == 0)
            intersect = intersect .* probMatrix(i,j).*log(probMatrix(i,j));
        else
            % if any of the features have 0 intersection, the rest are
            % disregarded
            intersect = 0;
            flagIntersect = 1;
        end
        
        if(AnotB(j)~=0 && flagA == 0)
            exclusiveA = exclusiveA .* probMatrix(i,j).*log(probMatrix(i,j));
        else
            exclusiveA = 0;
            flagA = 1;
        end
        
        if(BnotA(j)~=0 && flagB == 0)
            exclusiveB = exclusiveB .* probMatrix(i,j).*log(probMatrix(i,j));
        else
            exclusiveB = 0;
            flagB = 1;            
        end
    end
    sim = sim - intersect + exclusiveA./2 + exclusiveB./2
end