Implementation for custom distances?

[ninamiolane]

Hi, very nice package!

Is it possible to use it with a distance correlation coming from a custom notion of distance between Xi/Xj or Yi/Yj (i.e. not necessarily an exponent of the Euclidean distance?). If not, would you know of an implementation available to do so?

Thank you!

[vnmabus]

Currently it is not possible. It could be done, at least for the naive algorithm, propagating the distance function until the point where pairwise_distances is called, calling the distance function instead. Currently I have no time for this, but I would review a PR if someone takes the time.

You have several options:

• Use the R package energy, from the original authors of distance correlation. They allow you to pass a precomputed distance matrix.
• Download this package and replace pairwise_distances by your function (the ugly, one-time solution, but may work for you).
• Take a little more time to implement a robust solution (passing a distance callable through functions that need it) and propose a PR. I will review it, and if it is ok you would have help others with the same problem.

[ninamiolane]

Thank you very much for your quick answer! I saw the energy package, but I was looking for something in Python.

Thanks also for the indication on how to implement it, I'll see if I can find time and if so will submit a PR.

[vnmabus]

Before performing any work, I advise you to research if that is a sound procedure, that is, if the good properties of distance correlation are conserved if the distance is not Euclidean. I am not aware if that is the case, and I would prefer not to offer any option that can mislead the users of the package.