def geweke_joint_distribution_test(trace_1, trace_2, g):
'''
Compare whether the trace of two samples have converged to the same stationary distribution.
Returns :
p : p-value two simulators are drawing from same joint distribution.
'''
g_1 = [g(x) for x in trace_1]
g_2 = [g(x) for x in trace_2]
z = two_sample_z_score(g_1, g_2)
return 1 - inverse_normal_cdf(z)
def geweke_convergence_test(trace, first=0.1, last=0.5):
# Filter out invalid intervals
if first + last >= 1:
raise ValueError("Length of intervals must sum to <= 1. Values {0} and {1} sum to {2}.".format(first,
last,
first + last))
# Calculate starting indices
end_of_first_slice = int(len(trace) * first)
start_of_last_slice = int(len(trace) * (1 - last))
# Calculate slices
first_slice = trace[:end_of_first_slice]
last_slice = trace[start_of_last_slice:]
z = two_sample_z_score(first_slice, last_slice)
return 1 - inverse_normal_cdf(z)
alpha, partition = draw_from_prior(base_measure, size)
observed_p = partition.item_values[0].x
params = [param.x for param in partition.item_values]
data = draw_data(params, n)
sampler = Sampler(base_measure, cluster_density)
posterior = []
for _ in range(num_iters):
alpha, partition = sampler.sample(alpha, partition, data)
posterior.append(partition.item_values[0].x)
q = 0
for value in posterior:
if observed_p > value:
q += 1
q = q / len(posterior)
print q, inverse_normal_cdf(q), observed_p, mean(posterior)
test_stat.append(inverse_normal_cdf(q)**2)
print 1 - chi_square_cdf(sum(test_stat), len(test_stat))