Is there any assumption made when calculating mutualinfo(x, y) if one of vector x,y is binary?
In my data, y is binary and I tried several different x. Surprisingly, several different x give the same mutualinfo to the last digit of double type.
I think some assumption must be made on estimating kernel density. Could author give some insight?
Maybe I'm missing something, but I think this code produces weird behavior with vectors of low values, e.g. returning entropy of 0 for [1:10] * .0001.
Run this code and see for yourself:
a = ;
for i = 1:20e4
a(i) = entropy([1:10] .* i/10e4);