Fraser-Swinney lag estimate

This is a Mex wrapper for the original script minfo.c written by Eric Weeks (see references). The script estimates the mutual-information between an input time-series and its lagged versions for several lags. This is typically used to select the optimal lag for phase-space reconstruction. The method is published in:

Independent coordinates for strange attractors from mutual information
A. Fraser, H. Swinney, PRA 1986 (doi: 10.1103/PhysRevA.33.1134)

Compiling

Simply run

    mex -largeArrayDims -O -lut milag.cpp

from the command-line in Matlab. Note the linking to -lut for keyboard-interruption support.

Usage

The script takes one to three arguments:

X -- an Ntimes x Nchannels matrix of time-series in each column
lag_max (optional) -- the maximum lag to consider (see DEFAULT_LAG_MAX)
nbins (optional) -- number of bins

The computation can be interrupted from the Matlab command-line using Ctrl+C.

Notes

It is actually unclear to me how nbins affects the results, and I cannot see how this corresponds to the number of bins (which assumes that mutual information is estimated from a histogram, which is really not immediate from the code). I would advise not to use this input.

The original author (Eric Weeks) released his code in the public domain, but Dmytro Lituiev copyrighted his version despite a minimal contribution. I'm releasing this in the public domain, please don't hesitate to drop me a line for copyright concerns.

Output

Returns a (lag_max+1) x Nchannels matrix with the MI estimates for each lag and channel.

References

Andrew M. Fraser and Harry L. Swinney. "Independent coordinates for strange attractors from mutual information", Phys. Rev. A 33 (1986) 1134-1140.
Eric Weeks' page with the description of the modified algorithm

Contributors

Jonathan Hadida (2016), University of Oxford
Dmytro S. Lituiev (2013), University of Zurich
Eric Weeks C script (1997)