Greg Ver Steeg

It could be a question of which quantities it expects to be lists of vectors, and which not. I'd try modifying where the brackets are, e.g. this: ee.micd(cont.iloc[:,[1]].values.tolist(),disc.iloc[:,1].values.tolist())) That way,...

(I'm pretty sure discrete expects just a single discrete quantity, not a vector, but continuous does expect a vector.)

That makes sense. If X is continuous and Z is discrete, then I(X;Z) = H(Z) - H(Z|X)

I'm not sure I understand the example. What is the x-axis, and the difference between blue and red? While mutual information should never be negative, the estimator can be negative....

To add another point: the shuffle test is trying to estimate the bias of the mutual information estimator. It does so by shuffling data (to get a case that should...

Thanks for the comment. Sorry, I wrote this before I was very savvy about python programming. Even better would be a "pip install npeet". I will try to get someone...

Right! If these are continuous random variables, then there is another wrinkle. H(F|X) actually goes to negative infinity, and the mutual information becomes infinite (for a noiseless function). This is...

Those dimensionalities are a little high for the nearest neighbor based estimators. It might work if the signal is intrinsically low dimensional. If not, there have been some recent ideas...

I've heard from others that MINE is actually not that stable, so it may not just be you. You can always do dimensionality reduction to get a LOWER bound on...

Some of the issues with MINE are discussed in this paper: http://bayesiandeeplearning.org/2018/papers/136.pdf They also suggest a way to get more stable estimates, but I didn't see code posted.