def moyenneMMD_MC(methodeMMD,
kernel,
law_p,
law_q,
m,
n,
repeat=100,
verbose=1):
mmd_list = np.zeros(repeat)
for i in range(repeat):
if verbose > 0:
print(i / repeat * 100, '% achevé', end='\r', flush=True)
X_ = law_p(m)
Y_ = law_q(n)
MMD = methodeMMD(kernel)
MMD.fit(Y_, X_)
mmd_list[i] = MMD.MMD
return mmd_list.mean()
def moyenneMMD_Bt(methodeMMD,
kernel,
X,
Y,
n_boot,
m_boot,
repeat=100,
verbose=1):
mmd_list = np.zeros(repeat)
for i in range(repeat):
if verbose > 0:
print(i / repeat * 100, '% achevé', end='\r', flush=True)
X_ = utils.bootstrap(X, m_boot)
Y_ = utils.bootstrap(Y, n_boot)
MMD = methodeMMD(kernel)
MMD.fit(Y_, X_)
mmd_list[i] = MMD.MMD
return mmd_list.mean()
def sampleMMD_MC(methodeMMD,
kernel,
law_p,
law_q,
m,
n,
finalSampleSize=1000,
verbose=1):
Liste_mmdl = []
for _ in range(finalSampleSize):
if verbose > 0:
print(_ / finalSampleSize * 100, '% achevé', end='\r', flush=True)
X = law_p(m)
Y = law_q(n)
MMD = methodeMMD(kernel)
MMD.fit(Y, X)
Liste_mmdl.append(MMD.MMD)
return np.r_[Liste_mmdl]
# Compute U-statistics and bootstrap intervals
ksd_stats, _ = ksd_est([kappa_vals])
ksd_boot_list, ksd_thres_list = ksd_boot([kappa_vals])
ksd_pvals = ksd_pvalue(ksd_boot_list, ksd_stats)
ksd_stat = ksd_stats[0]
ksd_thres = ksd_thres_list[0]
ksd_pval = ksd_pvals[0]
ksd_pred = 1 * (ksd_stat > ksd_thres) # 0 for p, 1 for q
# ------------------------- Perform MMD test ------------------------- #
print "Performing MMD test ..."
mmd = MMD(kernel_fun=wl_kernel)
mmd_stat, mmd_thres, mmd_pval, _ = mmd.perform_test(samples_p, samples_q)
mmd_pred = 1 * (mmd_stat > mmd_thres) # 0 for p, 1 for q
res = {'n': n, 'param0': [rho0, theta0, tau0], 'param': [rho, theta, tau],
'true_dist': true_dist,
'ksd_stat': ksd_stat, 'ksd_thres': ksd_thres,
'ksd_pval': ksd_pval, 'ksd_pred': ksd_pred,
'mmd_stat': mmd_stat, 'mmd_thres': mmd_thres,
'mmd_pval': mmd_pval, 'mmd_pred': mmd_pred}
pckl_write(res, res_dir + "pow-ergm-n%d-d%d-rho%.3f-theta%.3f-tau%.3f-seed%d.res" %
(n, d, rho, theta, tau, seed))
print 'Finished!'