corrcorr.py

import numpy as np
import scipy.stats as st

nperm = 100
mix = 0.1
score = np.zeros((5, 17))
nparticipants = [int(i) for i in np.logspace(1, 5, 5)]
for ipar, npar in enumerate(nparticipants):
    for ivar, nvar in enumerate(range(3, 20)):
        accu = 0
        for perm in range(nperm):
            X1 = np.random.randn(nvar, npar)
            X2 = np.random.randn(nvar, npar)
            X1[1, :] = mix * X1[2, :] + (1 - mix) * np.random.randn(1, npar)
            X2[1, :] = mix * X2[2, :] + (1 - mix) * np.random.randn(1, npar)
            indexu = np.argwhere(np.triu(np.ones((nvar, nvar)), 1))
            C1 = np.triu(np.corrcoef(X1))[indexu[:, 0], indexu[:, 1]]
            C2 = np.triu(np.corrcoef(X2))[indexu[:, 0], indexu[:, 1]]
            R, p = st.pearsonr(C1, C2)
            if p<0.05:
                accu +=1
        score[ipar, ivar] = accu/nperm

plt.imshow(score)
plt.yticks(range(len(nparticipants)), nparticipants)
plt.ylabel("Sample size")
plt.xticks(range(17), range(3, 20))
plt.xlabel("Variables number")
plt.colorbar()
plt.show()