Preferred direction of flow

[kahaaga]

Implement a measure for the preferred all bivariate directional association measures.

• [ ] Transfer entropy (TE). This will be a wrapper function that computes normalised TE (NTE) in both directions.
• [ ] Convergent cross mapping (CCM). Requires normalisation of the CCM prediction skill to the ability of the target variable to predict itself.

This was used for transfer entropy in Gourévitch, B., & Eggermont, J. J. (2007). Evaluating information transfer between auditory cortical neurons. Journal of Neurophysiology, 97(3), 2533–2543.

[kahaaga]

This could be done by implementing a PreferredDirectionTest{CT} where {CT <: CausalityTest}. Analogously to PredictiveAsymmetryTest, this takes any causality test and automatically performs the causality tests both from source to target and target to source. It then takes the difference between the statistics in both directions, and returns a single statistic saying something about the direction.

There might be need for normalisation for some of the causality statistics.

[kahaaga]

The comment above refers to old, deleted code and can be disregarded.

On a new note: this feature should still be implemented, since it is so widely used. However, it is only applicable to directional measures. We can use the type parameter system to indicate which measures are directional and which are not, so we can make a unified API for this.

EDIT: a preferred direction measure is also used for the rank-statistic measures such as the S-measure & friends (#35 )