def test_prob_with_group_events_ndims(self):
with self.get_session():
distrib = Normal(np.asarray(0., dtype=np.float32),
np.asarray([1.0, 2.0, 3.0], dtype=np.float32),
group_event_ndims=1)
observed = np.asarray([[-1., 1., 2.], [0., 0., 0.]])
t = StochasticTensor(distrib, observed=observed)
np.testing.assert_allclose(
t.prob(group_event_ndims=0).eval(),
distrib.prob(t, group_event_ndims=0).eval())
np.testing.assert_allclose(
t.prob(group_event_ndims=1).eval(),
distrib.prob(t, group_event_ndims=1).eval())
def test_prob_and_log_prob(self):
with self.get_session():
distrib = Normal(np.asarray(0., dtype=np.float32),
np.asarray([1.0, 2.0, 3.0], dtype=np.float32))
observed = np.arange(24, dtype=np.float32).reshape([4, 2, 3])
t = StochasticTensor(distrib, observed=observed)
np.testing.assert_almost_equal(t.log_prob().eval(),
distrib.log_prob(observed).eval())
np.testing.assert_almost_equal(t.prob().eval(),
distrib.prob(observed).eval())
def test_specialized_prob_method(self):
class MyNormal(Normal):
@property
def has_specialized_prob_method(self):
return True
with self.get_session():
distrib = MyNormal(np.asarray(0., dtype=np.float32),
np.asarray([1.0, 2.0, 3.0], dtype=np.float32))
observed = np.arange(24, dtype=np.float32).reshape([4, 2, 3])
t = StochasticTensor(distrib, observed=observed)
t.distribution._has_specialized_prob_method = True
np.testing.assert_almost_equal(t.prob().eval(),
distrib.prob(observed).eval())