Mutual information for count data

[rvinas]

Hello,

Thank you for the nice library!

I would like to compute the mutual information between two count random variables. I have read the documentation and tried the ML, James-Stein, and Good-Turing estimators. Unfortunately, when the alphabets are large (i.e. range of possible counts), the James-Stein and Good-Turing estimators lead to negative mutual information.

Are these approaches suitable for estimating the mutual information for count random variables? Any advice or recommendations would be greatly appreciated.

Thank you and best wishes, Ramon