pySINGLE

Python implementation of SINGLE algorithm. Full details can be found [here] (http://arxiv.org/pdf/1310.3863.pdf) or alternatively [here] (http://mirrors.softliste.de/cran/web/packages/SINGLE/vignettes/vignette.pdf)

Requires the following (non-standard) python modules:

• spams: an optimization toolbox for solving various sparse estimation problems. Details can be found [here] (http://spams-devel.gforge.inria.fr/)

Notes:

The SINGLE class has the following methods (see example below):

1. fit_radius: this is used to estimate the radius of the Gaussian kernel, h.
2. tune_params: this function is used to estimate sparsity, l1, and temporal homogeneity, l2, parameters
3. fit: Once parameters h, l1 and l2 have been estimated this function can be used to estimate an array of sparse precision matrices using the SINGLE algorithm.
4. plot: used to plot resulting partial correlations

Installation:

Install using pip:

pip install pySINGLE

or using easy_install:

easy_install pySINGLE

Example:

We provide simulated data in the Sample Data folder.

import numpy, pandas
from pySINGLE.SINGLE import SINGLE

# read in data:
data = numpy.array(pandas.read_csv('pySINGLE/Sample Data/biggerdata.csv'))

# establish instance of SINGLE class:
S = SINGLE(data=data)

# estimate width of Gaussian kernel, h:
S.fit_radius(h_vals=[5,10,15], samples=None)

# estimate sparsity and temporal homogeneity parameters:
S.tune_params()

# finally, we can fit the model according to the estimated parameters:
S.fit()

# and plot partial correlations between nodes 1,2,6 and 7 (image below)
S.plot(index=[1,2,6,7], ncol_=2)