Average mutual information

Very fast implementation of average mutual information.

Usage: [v,lag]=ami(x,y,lag)

Calculates the mutual average information of x and y with a possible lag.


v is the average mutual information. (relative units see below)
x & y is the time series. (column vectors)
lag is a vector of time lags.

(A peak in V for lag>0 means y is leading x.)

v is given as how many bits x and y has in common relative to how
many bits is needed for the internally binned representation of x or y.
This is done to make the result close to independent bin size.

For optimal binning: transform x and y into percentiles prior to running
ami. See boxpdf at matlab central.

http://www.imt.liu.se/~magnus/cca/tutorial/node16.html

Aslak Grinsted feb2006
http://www.glaciology.net
(Inspired by mai.m by Alexandros Leontitsis)