def f_nll(n, m):
P = m.edge_probabilities(n)
w = P / (1.0 - P)
A = np.array(n.adjacency_matrix())
r, c = A.sum(1), A.sum(0)
z = approximate_from_margins_weights(r, c, w, T,
sort_by_wopt_var = False)
logkappa, logcvsq = log_partition_is(z, cvsq = True)
print 'est. cv^2 = %.2f (T = %d)' % (np.exp(logcvsq), T)
return (logkappa - np.sum(np.log(w[A])))
A = a.array
r = A.sum(1)
c = A.sum(0)
print r
print c
X = a.edge_covariates['x'].matrix()
for l, theta_l in enumerate(theta_vec):
print l
logit_P_l = theta_l * X
w_l = np.exp(logit_P_l)
cmle_a_vec[l] = -cond_a_nll(A, w_l)
z = cond_a_sample(r, c, w_l, T_grid)
logkappa, logcvsq = log_partition_is(z, cvsq=True)
cvsq = np.exp(logcvsq)
logkappa_cvsq[l] = cvsq
print 'est. cv^2 = %.2f (T_grid = %d)' % (cvsq, T_grid)
cmle_is_vec[l] = np.sum(np.log(w_l[A])) - logkappa
print 'CMLE-A: %.2f' % theta_vec[np.argmax(cmle_a_vec)]
print 'CMLE-IS: %.2f' % theta_vec[np.argmax(cmle_is_vec)]
plt.figure()
plt.plot(theta_vec, cmle_a_vec)
plt.plot(theta_vec, cmle_is_vec)
plt.figure()
plt.plot(theta_vec, logkappa_cvsq)
A = a.array
r = A.sum(1)
c = A.sum(0)
print r
print c
X = a.edge_covariates['x'].matrix()
for l, theta_l in enumerate(theta_vec):
print l
logit_P_l = theta_l * X
w_l = np.exp(logit_P_l)
cmle_a_vec[l] = -cond_a_nll(A, w_l)
z = cond_a_sample(r, c, w_l, T_grid)
logkappa, logcvsq = log_partition_is(z, cvsq = True)
cvsq = np.exp(logcvsq)
logkappa_cvsq[l] = cvsq
print 'est. cv^2 = %.2f (T_grid = %d)' % (cvsq, T_grid)
cmle_is_vec[l] = np.sum(np.log(w_l[A])) - logkappa
print 'CMLE-A: %.2f' % theta_vec[np.argmax(cmle_a_vec)]
print 'CMLE-IS: %.2f' % theta_vec[np.argmax(cmle_is_vec)]
plt.figure()
plt.plot(theta_vec, cmle_a_vec)
plt.plot(theta_vec, cmle_is_vec)
plt.figure()
plt.plot(theta_vec, logkappa_cvsq)
