Algorithm 2:

kdq bayes KL (for homogeneity of adjacent windows)

data: a binary array of [Ndatapoints by Nvariables];
data.sh ← a time-shuffled copy of data (null distribution);
γ: sets the number of data points within a window;
stepsize: sets the number of data points between comparisons;
α sets the value of the parameters of the Dirichlet prior;
T ← Construct Binary kdq-tree (data, splitmin);
i ← 1;
while not at the end of the data do
 Load adjacent samples W1 and W2 of γ samples from data;
 Likewise load W1.sh and W2.sh from data.sh;
 n1 ← multinomial sample generated by filtering W1 through T;
 n2 ← multinomial sample generated by filtering W2 through T;
 n1.sh ← multinomial sample generated by filtering W1.sh through T;
 n2.sh ← multinomial sample generated by filtering W2.sh through T;
 KLs[i+γ-1] ← bayes KL(n1, n2, α);
 KLs.sh[i+γ−1] ← bayes KL(n1.sh, n2.sh, α);
 Slide windows W1, W2, W1.sh and W2.sh by stepsize;
 i ← i+1;
end while
null.mode ← mode(KLs.sh[γ to end- γ by stepsize]);
null.std ← standard deviation(KLs.sh[γ to end- γ by stepsize]);