def testMixtureTargetLogProb(self):
seed = test_util.test_seed()
n = 4
mu = np.ones(n) * (1. / 2)
w = 0.1
proposal = tfd.Sample(tfd.Normal(0., 10.), sample_shape=n)
init_state = proposal.sample(5000, seed=seed)
likelihood_dist = tfd.MixtureSameFamily(
mixture_distribution=tfd.Categorical(probs=[w, 1. - w]),
components_distribution=tfd.MultivariateNormalDiag(
loc=np.asarray([mu, -mu]).astype(np.float32),
scale_identity_multiplier=[.1, .2]))
# Uniform prior
init_log_prob = tf.zeros_like(proposal.log_prob(init_state))
[n_stage, final_state,
_] = tfp.experimental.mcmc.sample_sequential_monte_carlo(
lambda x: init_log_prob,
likelihood_dist.log_prob,
init_state,
max_num_steps=50,
parallel_iterations=1,
seed=seed)
self.assertTrue(self.evaluate(n_stage), 15)
self.evaluate(
st.kolmogorov_smirnov_distance_two_sample(
final_state, likelihood_dist.sample(5000, seed=seed)))
def test_kolmogorov_smirnov_distance_two_sample(self, dtype):
del dtype
samples1 = np.array([[1, 1, 2, 2], [1, 2, 3, 4], [1, 3, 3, 3]])
samples2 = np.array([[1, 2, 2, 1, 1, 2], [4, 4, 3, 2, 1, 1],
[2, 2, 2, 1, -4, 7]])
# Unlike empirical_cdfs, the samples Tensors must come in iid across the
# leading dimension.
obtained = self.evaluate(
st.kolmogorov_smirnov_distance_two_sample(
tf.transpose(a=samples1), tf.transpose(a=samples2)))
self.assertAllClose([0.0, 1 / 12., 0.75 - 1 / 6.], obtained, atol=1e-7)
