def testSingleTensor(self, bijector, dtype):
if not tf.executing_eagerly():
return
base_mean = tf.convert_to_tensor(value=[1., 0], dtype=dtype)
base_cov = tf.convert_to_tensor(value=[[1, 0.5], [0.5, 1]],
dtype=dtype)
base_dist = tfd.MultivariateNormalFullCovariance(
loc=base_mean, covariance_matrix=base_cov)
target_dist = bijector(base_dist)
def debug_fn(*args):
del args
debug_fn.count += 1
debug_fn.count = 0
kernel = neutra.NeuTra(
target_log_prob_fn=target_dist.log_prob,
state_shape=2,
num_step_size_adaptation_steps=800,
num_train_steps=1000,
train_batch_size=64,
learning_rate=tf.convert_to_tensor(value=1e-2, dtype=dtype),
seed=tfp_test_util.test_seed(),
train_debug_fn=debug_fn,
unconstraining_bijector=bijector,
)
chain = tfp.mcmc.sample_chain(num_results=1000,
num_burnin_steps=1000,
current_state=tf.zeros([16, 2], dtype),
kernel=kernel,
trace_fn=None,
parallel_iterations=1)
self.assertEqual(1000, debug_fn.count)
sample_mean = tf.reduce_mean(input_tensor=chain, axis=[0, 1])
sample_cov = tfp.stats.covariance(chain, sample_axis=[0, 1])
true_samples = target_dist.sample(4096, seed=tfp_test_util.test_seed())
true_mean = tf.reduce_mean(input_tensor=true_samples, axis=0)
true_cov = tfp.stats.covariance(chain, sample_axis=[0, 1])
self.assertAllClose(true_mean, sample_mean, rtol=0.1, atol=0.1)
self.assertAllClose(true_cov, sample_cov, rtol=0.1, atol=0.1)
def testNested(self, bijector):
if not tf.executing_eagerly():
return
base_mean = tf.constant([1., 0])
base_cov = tf.constant([[1, 0.5], [0.5, 1]])
dist_2d = tfd.MultivariateNormalFullCovariance(
loc=base_mean, covariance_matrix=base_cov)
dist_4d = tfd.MultivariateNormalDiag(scale_diag=tf.ones(4))
target_dist = tfd.JointDistributionSequential([
bijector(dist_2d),
tfb.Reshape([2, 2])(dist_4d),
])
kernel = neutra.NeuTra(
target_log_prob_fn=lambda x, y: target_dist.log_prob((x, y)),
state_shape=target_dist.event_shape,
num_step_size_adaptation_steps=800,
num_train_steps=1000,
train_batch_size=64,
seed=tfp_test_util.test_seed(),
unconstraining_bijector=[bijector, tfb.Identity()],
)
chain_2d, chain_4d = tfp.mcmc.sample_chain(
num_results=1000,
num_burnin_steps=1000,
current_state=tf.nest.map_structure(
lambda s: tf.zeros([16] + s.as_list()),
target_dist.event_shape),
kernel=kernel,
trace_fn=None,
parallel_iterations=1)
sample_mean_2d = tf.reduce_mean(input_tensor=chain_2d, axis=[0, 1])
sample_mean_4d = tf.reduce_mean(input_tensor=chain_4d, axis=[0, 1])
true_samples_2d, true_samples_4d = target_dist.sample(
4096, seed=tfp_test_util.test_seed())
true_mean_2d = tf.reduce_mean(input_tensor=true_samples_2d, axis=0)
true_mean_4d = tf.reduce_mean(input_tensor=true_samples_4d, axis=0)
self.assertAllClose(true_mean_2d, sample_mean_2d, rtol=0.1, atol=0.1)
self.assertAllClose(true_mean_4d, sample_mean_4d, rtol=0.1, atol=0.1)