def test_Normal(self):
with BayesianNet():
mean = tf.zeros([2, 3])
logstd = tf.zeros([2, 3])
std = tf.exp(logstd)
n_samples = tf.placeholder(tf.int32, shape=[])
group_event_ndims = tf.placeholder(tf.int32, shape=[])
a = Normal('a',
mean,
logstd=logstd,
n_samples=n_samples,
group_event_ndims=group_event_ndims)
b = Normal('b',
mean,
std=std,
n_samples=n_samples,
group_event_ndims=group_event_ndims)
for st in [a, b]:
sample_ops = set(get_backward_ops(st.tensor))
for i in [mean, logstd, n_samples]:
self.assertTrue(i.op in sample_ops)
log_p = st.log_prob(np.ones([2, 3]))
log_p_ops = set(get_backward_ops(log_p))
for i in [mean, logstd, group_event_ndims]:
self.assertTrue(i.op in log_p_ops)
def test_Discrete(self):
with BayesianNet():
logits = tf.zeros([2, 3])
n_samples = tf.placeholder(tf.int32, shape=())
group_event_ndims = tf.placeholder(tf.int32, shape=[])
a = Categorical('a', logits, n_samples, group_event_ndims)
sample_ops = set(get_backward_ops(a.tensor))
for i in [logits, n_samples]:
self.assertTrue(i.op in sample_ops)
log_p = a.log_prob(np.array([0, 1]))
log_p_ops = set(get_backward_ops(log_p))
for i in [logits, group_event_ndims]:
self.assertTrue(i.op in log_p_ops)
def test_Bernoulli(self):
with BayesianNet():
logits = tf.zeros([2, 3])
n_samples = tf.placeholder(tf.int32, shape=[])
group_event_ndims = tf.placeholder(tf.int32, shape=[])
a = Bernoulli('a', logits, n_samples, group_event_ndims)
sample_ops = set(get_backward_ops(a.tensor))
for i in [logits, n_samples]:
self.assertTrue(i.op in sample_ops)
log_p = a.log_prob(np.ones([2, 3]))
log_p_ops = set(get_backward_ops(log_p))
for i in [logits, group_event_ndims]:
self.assertTrue(i.op in log_p_ops)
def test_Dirichlet(self):
with BayesianNet():
alpha = tf.ones([2, 3])
n_samples = tf.placeholder(tf.int32, shape=[])
group_event_ndims = tf.placeholder(tf.int32, shape=[])
a = Dirichlet('a', alpha, n_samples, group_event_ndims)
sample_ops = set(get_backward_ops(a.tensor))
for i in [alpha, n_samples]:
self.assertTrue(i.op in sample_ops)
log_p = a.log_prob(np.array([[0.2, 0.3, 0.5], [0.1, 0.7, 0.2]]))
log_p_ops = set(get_backward_ops(log_p))
for i in [alpha, group_event_ndims]:
self.assertTrue(i.op in log_p_ops)
def test_OnehotCategorical(self):
with BayesianNet():
logits = tf.ones([2, 3])
n_samples = tf.placeholder(tf.int32, shape=[])
group_event_ndims = tf.placeholder(tf.int32, shape=[])
a = OnehotCategorical('a', logits, n_samples, group_event_ndims)
sample_ops = set(get_backward_ops(a.tensor))
for i in [logits, n_samples]:
self.assertTrue(i.op in sample_ops)
log_p = a.log_prob(tf.one_hot([0, 2], 3, dtype=tf.int32))
log_p_ops = set(get_backward_ops(log_p))
for i in [logits, group_event_ndims]:
self.assertTrue(i.op in log_p_ops)
def test_Poisson(self):
with BayesianNet():
rate = tf.ones([2, 3])
n_samples = tf.placeholder(tf.int32, shape=[])
group_event_ndims = tf.placeholder(tf.int32, shape=[])
a = Poisson('a', rate, n_samples, group_event_ndims)
sample_ops = set(get_backward_ops(a.tensor))
for i in [rate, n_samples]:
self.assertTrue(i.op in sample_ops)
log_p = a.log_prob(np.ones([2, 3], dtype=np.int32))
log_p_ops = set(get_backward_ops(log_p))
for i in [rate, group_event_ndims]:
self.assertTrue(i.op in log_p_ops)
def test_Beta(self):
with BayesianNet():
alpha = tf.ones([2, 3])
beta = tf.ones([2, 3])
n_samples = tf.placeholder(tf.int32, shape=[])
group_event_ndims = tf.placeholder(tf.int32, shape=[])
a = Beta('a', alpha, beta, n_samples, group_event_ndims)
sample_ops = set(get_backward_ops(a.tensor))
for i in [alpha, beta, n_samples]:
self.assertTrue(i.op in sample_ops)
log_p = a.log_prob(np.ones([2, 3]) * 0.5)
log_p_ops = set(get_backward_ops(log_p))
for i in [alpha, beta, group_event_ndims]:
self.assertTrue(i.op in log_p_ops)
def test_Uniform(self):
with BayesianNet():
minval = tf.zeros([2, 3])
maxval = tf.ones([2, 3])
n_samples = tf.placeholder(tf.int32, shape=[])
group_event_ndims = tf.placeholder(tf.int32, shape=[])
a = Uniform('a', minval, maxval, n_samples, group_event_ndims)
sample_ops = set(get_backward_ops(a.tensor))
for i in [minval, maxval, n_samples]:
self.assertTrue(i.op in sample_ops)
log_p = a.log_prob(np.zeros([2, 3]))
log_p_ops = set(get_backward_ops(log_p))
for i in [minval, maxval, group_event_ndims]:
self.assertTrue(i.op in log_p_ops)
def test_Laplace(self):
with BayesianNet():
loc = tf.zeros([2, 3])
scale = tf.ones([2, 3])
n_samples = tf.placeholder(tf.int32, shape=[])
group_ndims = tf.placeholder(tf.int32, shape=[])
a = Laplace('a', loc, scale, n_samples, group_ndims)
sample_ops = set(get_backward_ops(a.tensor))
for i in [loc, scale, n_samples]:
self.assertTrue(i.op in sample_ops)
log_p = a.log_prob(np.ones([2, 3]))
log_p_ops = set(get_backward_ops(log_p))
for i in [loc, scale, group_ndims]:
self.assertTrue(i.op in log_p_ops)
def test_Multinomial(self):
with BayesianNet():
logits = tf.ones([2, 3])
n_experiments = tf.placeholder(tf.int32, shape=[])
n_samples = tf.placeholder(tf.int32, shape=[])
group_ndims = tf.placeholder(tf.int32, shape=[])
a = Multinomial('a', logits, n_experiments, n_samples, group_ndims)
sample_ops = set(get_backward_ops(a.tensor))
for i in [logits, n_experiments, n_samples]:
self.assertTrue(i.op in sample_ops)
log_p = a.log_prob(np.ones([2, 3], dtype=np.int32))
log_p_ops = set(get_backward_ops(log_p))
for i in [logits, n_experiments, group_ndims]:
self.assertTrue(i.op in log_p_ops)
def test_ExpConcrete(self):
with BayesianNet():
logits = tf.zeros([2, 3])
tau = tf.ones([])
n_samples = tf.placeholder(tf.int32, shape=[])
group_ndims = tf.placeholder(tf.int32, shape=[])
a = ExpConcrete('a', tau, logits, n_samples, group_ndims)
sample_ops = set(get_backward_ops(a.tensor))
for i in [logits, tau, n_samples]:
self.assertTrue(i.op in sample_ops)
log_p = a.log_prob(np.ones([2, 3]))
log_p_ops = set(get_backward_ops(log_p))
for i in [logits, tau, group_ndims]:
self.assertTrue(i.op in log_p_ops)
self.assertEqual(a.get_shape()[1:], logits.get_shape())
def test_InverseGamma(self):
with BayesianNet():
alpha = tf.ones([2, 3])
beta = tf.ones([2, 3])
n_samples = tf.placeholder(tf.int32, shape=[])
group_ndims = tf.placeholder(tf.int32, shape=[])
a = InverseGamma('a', alpha, beta, n_samples, group_ndims)
sample_ops = set(get_backward_ops(a.tensor))
for i in [alpha, beta, n_samples]:
self.assertTrue(i.op in sample_ops)
log_p = a.log_prob(np.ones([2, 3]))
log_p_ops = set(get_backward_ops(log_p))
for i in [alpha, beta, group_ndims]:
self.assertTrue(i.op in log_p_ops)
self.assertEqual(a.get_shape()[1:], alpha.get_shape())
def test_UnnormalizedMultinomial(self):
with BayesianNet():
logits = tf.ones([2, 3])
group_ndims = tf.placeholder(tf.int32, shape=[])
a = UnnormalizedMultinomial('a', logits, group_ndims=group_ndims)
log_p = a.log_prob(np.ones([2, 3], dtype=np.int32))
log_p_ops = set(get_backward_ops(log_p))
for i in [logits, group_ndims]:
self.assertTrue(i.op in log_p_ops)
def test_Implicit(self):
with BayesianNet() as model:
mean = tf.zeros([2, 3])
logstd = tf.zeros([2, 3])
n_samples = tf.placeholder(tf.int32, shape=[])
group_ndims = tf.placeholder(tf.int32, shape=[])
a = Normal('a',
mean,
logstd=logstd,
n_samples=n_samples,
group_ndims=group_ndims)
b = Implicit('b', a, value_shape=[])
sample_ops = set(get_backward_ops(b.tensor))
for i in [mean, logstd, n_samples]:
self.assertTrue(i.op in sample_ops)
self.assertEqual(a.get_shape().as_list(), b.get_shape().as_list())
(a_value, ), (b_value, ) = model.query(['a', 'b'], outputs=True)
# The ops are Squeeze(ExpandDims(a, 0))
self.assertTrue(a_value in b_value.op.inputs[0].op.inputs)