Ed Baskerville last update: 26 February 2015

NB: Usage instructions out of date but repo is published for the sake of reproducibility for paper.

This package contains a Python implementation of nonlinear time-series embedding methods.

Requires Python 2.7.x with numpy and scipy.

Recommended way to install Python:

• Anaconda distribution

After you have installed Anaconda and configured your PATH environment variable, make sure that

$ python --version

includes the word "Anaconda" in the description.

The script ccm.py automates the method of convergent cross-mapping (TODO: ref Sugihara et al.).

Run

ccm.py --help

to see detailed usage information.

Example:

ccm.py --n-cores 4 -E 4 -t 1 -L "5:50:5" -R 100 -C "x,y" -V "x:y,y:x" input.sqlite timeseries output.sqlite

will read time-series data from columns x and y (-C "x,y") in table timeseries in the SQLite database input.sqlite; test causality for x causing y as well as y causing x; and write output to the file output.sqlite; using these parameters:

• embedding dimension: 4 (-E 4)
• lag (tau): 1 (-t 1)
• library lengths: 5, 10, ..., 45, 50 (-L "5:50:5")
• replicate samples per library length: 100 (-R 100)

The output database will contain two tables:

• args: contains the arguments used to run the program
• results: contains analysis results.

The results table contains these columns:

• cause: the variable being tested as cause
• effect: the variable being tested as effect
• L: the library length being tested
• replicate_id: unique ID for this replicate
• corr: the correlation between the original "cause" time series and the version reconstructed from lagged vectors sampled from "effect".

TODO