How to handle nan values in refined multiscale entropy ?

[JulianKlug]

Given a sequence, how should nan values be handled for multiscale entropy?

Current behavior will raise an unhelpful error as r will be undefined: AssertionError: r: must be a positive value

This is due to the handling of nan in the used numpy functions, which will result in nan.

    if RadNew > 0:
        if RadNew == 1:
            Rnew = lambda x: np.std(x)
        elif RadNew == 2:
            Rnew = lambda x: np.var(x)
        elif RadNew == 3:
            Rnew = lambda x: np.mean(abs(x-np.mean(x)))
        elif RadNew == 4:
            Rnew = lambda x: np.median(abs(x-np.median(x)))   

as well as of the behavior of the function refined()

To solve this, what is considered best practice for the handling of nan in a signal before entropy computation?

(and maybe at the beginning of the function, nan's should be checked to have a more helpful error?)