pyMVGC 1.0.0

Multivariate Granger causality for time-series data in python

pyMVGC: Python Multivariate Granger causality

A Python implementation of multivariate Granger causality (MVGC) for time-series data, based on algorithms from Barnett & Seth's MVGC toolbox [1]. This method avoids explicit estimation of the reduced VAR via moving between autocovariance-sequence and regression-parameter representations of the VAR model. This reduces estimation error and imporoves statistical power and speed of estimation compared to the traditional estimation of separate full and reduced models.

Usage

pyMVGC can be downloaded from PyPI via pip:

$ pip install pyMVGC

Multivariate granger causality may be estimated in one line of code:

import pyMVGC as gc

# Multivariate Granger causality from one set of variables to another
ts_to_MVGC(data, to_idx = [0], from_idx = [1], p = 2)

#Pairwise granger causality between all variables in the system
ts_to_pMVGC(data, p = 2)

Optimal model order can be estimated from "AIC" or "BIC" scores:

estimate_model_order(data, maxLag = 30, criterion = "AIC")

Functions allowing movement between autocovariance and VAR representations, used in MVGC calculation, are also available:

Phi, resCov = gc.ts_to_var(data = data, p = p) # learn VAR from time series data.
autocov = gc.var_to_autocov(Phi = Phi, resCovMat = resCov) # learn autocovariance sequence from VAR.
Phi, resCov = gc.autocov_to_var(autocov = autocov)

GC = gc.autocov_to_MVGC(autocov, to_idx,from_idx) # get multivariate granger causality from from autocovarance sequence.
pGC = gc.autocov_to_pMVGC(autocov) # get pairwise granger causality from autocovariance sequnce.
    

See notebooks/tutorial.ipynb for a more detailed tutorial.

References

[1] L. Barnett and A. K. Seth, "The MVGC Multivariate Granger Causality Toolbox: A new approach to Granger-causal inference", J. Neurosci. Methods 223, pp 50-68, 2014.