def plot_mcmc_result(result, D1, D2):
plt.ioff()
samples = result.samples
accepted = result.accepted
log_pdf = result.log_pdf
avg_ess = result.posterior_statistics["avg_ess"]
min_ess = result.posterior_statistics["min_ess"]
time = result.time_taken_sampling
time_set_up = result.time_taken_set_up
# compute autocorrelation of the first dimension
acorrs = autocorr(samples[:, D1])
# visualise summary
plt.figure(figsize=(8, 12))
plt.subplot(521)
plt.plot(samples[:, D1])
plt.subplot(522)
plt.plot(samples[:, D2])
plt.subplot(523)
plt.hist(samples[:, D1])
plt.subplot(524)
plt.hist(samples[:, D2])
plt.subplot(525)
plt.plot(np.cumsum(accepted) / np.arange(1, len(samples) + 1))
plt.subplot(526)
plt.plot(acorrs)
plt.subplot(527)
plt.plot(samples[:, D1], samples[:, D2])
plt.subplot(528)
plt.plot(samples[:, D1], samples[:, D2], '.')
plt.subplot(529)
plt.plot(log_pdf, '-')
# print quantile summary
print("Average acceptance probability: %.2f" % np.mean(accepted))
print("Average ESS: %.2f" % avg_ess)
print("Minimum ESS: %.2f" % min_ess)
print("Average ESS/s: %.2f" % (avg_ess / time))
print("Minimum ESS/s: %.2f" % (min_ess / time))
print("Average ESS/s including set up: %.2f" % (avg_ess /
(time + time_set_up)))
plt.show()
def plot_diagnosis(agg, D):
plt.ioff()
samples = agg.result.samples
accepted = agg.result.accepted
avg_quantile_errors = agg.result.posterior_statistics["avg_quantile_error"]
avg_ess = agg.result.posterior_statistics["avg_ess"]
norm_of_mean = agg.result.posterior_statistics["norm_of_mean"]
time = agg.result.time_taken_sampling
time_set_up = agg.result.time_taken_set_up
# compute autocorrelation of the dimension with heavier tails
acorrs = autocorr(samples[:, 1])
# visualise summary
plt.figure(figsize=(8, 12))
plt.subplot(421)
plt.plot(samples[:, 0])
plt.subplot(422)
plt.plot(samples[:, 1])
plt.subplot(423)
plt.hist(samples[:, 0])
plt.subplot(424)
plt.hist(samples[:, 1])
plt.subplot(425)
plt.plot(np.cumsum(accepted) / np.arange(1, len(samples) + 1))
plt.subplot(426)
plt.plot(acorrs)
plt.subplot(427)
plt.plot(samples[:, 0], samples[:, 1])
plt.subplot(428)
plt.plot(samples[:, 0], samples[:, 1], '.')
# print quantile summary
print("Average quantile errors: %.2f" % avg_quantile_errors)
print("Average acceptance probability: %.2f" % np.mean(accepted))
print("Average ESS: %.2f" % avg_ess)
print("Average ESS/s: %.2f" % (avg_ess / time))
print("Average ESS/s including set up: %.2f" % (avg_ess /
(time + time_set_up)))
print("Average norm of mean: %.2f" % norm_of_mean)
plt.show()
# accept-reject
r = np.random.rand()
accepted[i] = r < acc_prob[i]
accepted_est[i] = r < acc_prob_est[i]
if accepted[i]:
q_current = proposals[i]
if accepted_est[i]:
q_current_est = proposals_est[i]
# update state
samples[i] = q_current
samples_est[i] = q_current_est
# compute autocorrelation of the dimension with heavier tails
acorrs = autocorr(samples[:, 1])
acorrs_est = autocorr(samples_est[:, 1])
# visualise summary
plt.figure(figsize=(8, 12))
plt.subplot(421)
plt.plot(samples[:, 0])
plt.subplot(422)
plt.plot(samples[:, 1])
plt.subplot(423)
plt.hist(samples[:, 0])
plt.subplot(424)
plt.hist(samples[:, 1])
plt.subplot(425)
plt.plot(np.cumsum(accepted) / np.arange(1, num_iterations + 1))
plt.subplot(426)
logger.info("Average acceptance probability: %.2f" %
np.mean(accepted))
logger.info("Average ESS: %.2f" % avg_ess)
logger.info("Minimum ESS: %.2f" % min_ess)
total_time_s = time + time_set_up
total_time_m = total_time_s / 60
logger.info("Total time: %.2f sec, %.2f min" %
(total_time_s, total_time_m))
logger.info("Average ESS/s: %.2f" % (avg_ess / total_time_s))
logger.info("Minimum ESS/s: %.2f" % (min_ess / total_time_s))
# store marginal samples
marginal_samples[alg_idx].append(samples)
# compute autocorrelation of the first dimension
acorrs[i, :] = autocorr(samples[:, 0])[:len_acorr]
med = np.median(acorrs, 0)
lower = np.percentile(acorrs, 20, 0)
upper = np.percentile(acorrs, 80, 0)
ax_acor.plot(med, "-", color=colors[alg_idx])
ax_acor.plot(lower, '--', color=colors[alg_idx])
ax_acor.plot(upper, '--', color=colors[alg_idx])
ax_acor.set_xlabel("Lag")
ax_acor.set_ylabel("Autocorrelation")
line1 = Line2D([0, 0], [0, 0], color=colors[0])
line2 = Line2D([0, 0], [0, 0], color=colors[1])
line3 = Line2D([0, 0], [0, 0], color=colors[2])
line4 = Line2D([0, 0], [0, 0], color=colors[3])