def test_real(self):
with self.test_session():
x = Normal(0.0, 1.0)
y = ed.transform(x)
self.assertIsInstance(y, Normal)
sample = y.sample(10, seed=1).eval()
self.assertSamplePosNeg(sample)
def test_simplex(self):
with self.test_session():
x = Dirichlet([1.1, 1.2, 1.3, 1.4])
y = ed.transform(x)
self.assertIsInstance(y, TransformedDistribution)
sample = y.sample(10, seed=1).eval()
self.assertSamplePosNeg(sample)
def test_nonnegative(self):
with self.test_session():
x = Gamma(1.0, 1.0)
y = ed.transform(x)
self.assertIsInstance(y, TransformedDistribution)
sample = y.sample(10, seed=1).eval()
self.assertSamplePosNeg(sample)
def test_unhandled_support(self):
with self.test_session():
FakeRV = namedtuple('FakeRV', ['support'])
x = FakeRV(support='rational')
with self.assertRaises(ValueError):
y = ed.transform(x)
def test_no_support(self):
with self.test_session():
x = DirichletProcess(1.0, Normal(0.0, 1.0))
with self.assertRaises(AttributeError):
y = ed.transform(x)
def test_multivariate_real(self):
with self.test_session():
x = MultivariateNormalDiag(tf.zeros(2), tf.ones(2))
y = ed.transform(x)
sample = y.sample(10, seed=1).eval()
self.assertSamplePosNeg(sample)
def test_kwargs(self):
with self.test_session():
x = Normal(-100.0, 1.0)
y = ed.transform(x, bijector=bijectors.Softplus())
sample = y.sample(10).eval()
self.assertTrue((sample >= 0.0).all())
def test_no_support(self):
with self.test_session():
x = PointMass(1.0)
with self.assertRaises(ValueError):
y = ed.transform(x)
probs=p,
concentrations=conc_param *
tf.ones(n_samples, dtype=np.float32),
sample_shape=n_samples,
value=tf.zeros(n_samples, dtype="float32"))
# INFERENCE
#qp = edward.models.BetaWithSoftplusConcentration(tf.Variable(1.), tf.Variable(1.))
qp = ed.models.Normal(loc=tf.get_variable("qp/loc", []),
scale=tf.nn.softplus(tf.get_variable("qp/scale", [])))
qconc = ed.models.Normal(loc=tf.get_variable("qconc/loc", []),
scale=tf.nn.softplus(
tf.get_variable("qconc/scale", [])))
inference = ed.KLqp({p: qp, conc_param: qconc}, data={x: x_data})
inference.run()
# PRINT RESULTS
qp_samples = ed.transform(
qp, bijectors.Invert(
inference.transformations[p].bijector)).sample(100).eval()
print("True prob success: {:.2f}, inferred {:.3f} +- {:.2f}".format(
p_true, qp_samples.mean(), np.sqrt(qp_samples.var())))
qconc_samples = ed.transform(
qconc, bijectors.Invert(
inference.transformations[conc_param].bijector)).sample(100).eval()
print("True concentration: {:.2f}, Inferred: {:.3f} +- {:.2f}".format(
true_conc, qconc_samples.mean(), np.sqrt(qconc_samples.var())))