Compute normalised TE (NTE)

[kahaaga]

Follow approach in Gourévitch, B., & Eggermont, J. J. (2007). Evaluating information transfer between auditory cortical neurons. Journal of Neurophysiology, 97(3), 2533–2543.

To compute NTE, subtract from the TE estimate the mean TE of a surrogate ensemble (computed on shuffled time series), and divide that by the conditional entropy H(target_future | target_past). NTE is then a number on [0, 1] and represents the fraction of information in the target time series not explained by its own past and explained by the past of the source time series.

The conditional entropy can easily be added to each of the existing estimators because they already compute TE as the sum of entropies of marginal distributions. Computing the full normalised TE requires running TE on a surrogate ensemble too, so we need a wrapper function for this.