Compute average mi and xcross across the data.

mutual_information.py

import numpy as np
import pyunicorn
import matplotlib.pyplot as pl
pl.ion()

data = np.load('all_fmri_data.npy')
mi_all = np.zeros((10, 90, 90))
xcorr_all = np.zeros((10, 90, 90))

from pyunicorn import funcnet

for subj in range(10):
    ca = funcnet.CouplingAnalysis(data[subj].T)
    mi_all[subj] = ca.mutual_information(tau_max=0, bins=4)[0]
    xcorr_all[subj] = ca.cross_correlation(tau_max=0)
    xcorr_all[subj] = (xcorr_all[subj] - np.min(xcorr_all[subj])) / (np.max(xcorr_all[subj]) - np.min(xcorr_all[subj]))
    assert xcorr_all.max() <= 1 and xcorr_all.min() >= 0., 'Not normalized properly...'

avg_mi = np.mean(mi_all, axis=0)
avg_mi = avg_mi - np.eye(len(avg_mi))
avg_xcorr = np.mean(xcorr_all, axis=0)
avg_xcorr = avg_xcorr - np.eye(len(avg_xcorr))

# read the channel names from the numpy file
channel_names = np.load('labels.npy')

# initializing the network
net = pyunicorn.Network(avg_mi)
#surr = net.ErdosRenyi(n_nodes=net.N, n_links=net.n_links)
link_weights = np.abs(avg_mi)
net.set_link_attribute(attribute_name="weight", values=link_weights)
net.set_node_attribute(attribute_name="label", values=channel_names)

print 'Saving the file...'
net.save('avg_mi.gml')  # save the network

# comute the network for average cross correlation
net2 = pyunicorn.Network(avg_xcorr)
#surr = net.ErdosRenyi(n_nodes=net.N, n_links=net.n_links)
link_weights = np.abs(avg_xcorr)
net2.set_link_attribute(attribute_name="weight", values=link_weights)
net2.set_node_attribute(attribute_name="label", values=channel_names)
net2.save('avg_xcorr.gml')  # save the network