Algorithm by: David Lopez-Paz, Philipp Hennig, and Bernhard Schoelkopf

Code by: Gary Doran

To install, run:

$ [sudo] python setup.py install

The RDC is a measure of nonlinear dependence between two (possibly multidimensional) variables. A full description of the algorithm is given in the 2013 paper by David Lopez-Paz, Philipp Hennig, and Bernhard Schoelkopf.

Given two NumPy arrays, x and y, the measure can be invoked as follows:

>>> from rdc import rdc
>>> print rdc(x, y)

If x and y are univariate, then they should be 1-D NumPy arrays; otherwise, then should be n-by-k arrays, where k is the number of dimensions and n is the number of examples. The two variables must have the same number of examples, but can have different numbers of features.

There are additional keyword parameters for rdc that correspond to parameters described in the paper. One new parameter is n, which is the number of times the RDC is computed with different random seeds to reduce variance in the estimation of the statistic. The median value across these n runs is returned.