def _make_posterior(std, weight_shape, full, suffix=None):
"""Check/make posterior."""
# We don't want a full-covariance on an intercept, check input_dim
if full and len(weight_shape) > 1:
post_W = gaus_posterior(weight_shape, std0=std, suffix=suffix)
else:
post_W = norm_posterior(weight_shape, std0=std, suffix=suffix)
assert _is_dim(post_W, weight_shape), \
"Posterior inconsistent dimension!"
return post_W
def _make_posterior(self, post_W, weight_shape):
"""Check/make posterior."""
if post_W is None:
# We don't want a full-covariance on an intercept, check input_dim
if self.full and len(weight_shape) > 1:
post_W = gaus_posterior(weight_shape, std0=self.std)
else:
post_W = norm_posterior(weight_shape, std0=self.std)
assert _is_dim(post_W, weight_shape), \
"Posterior inconsistent dimension!"
return post_W
Phi, KL = ab.Conv2DVariational(filters=D, kernel_size=(4, 4), **dists)(X_)
tc = tf.test.TestCase()
with tc.test_session():
tf.global_variables_initializer().run()
P = Phi.eval(feed_dict={x_: x})
assert P.shape == (S, N, height, width, D)
assert KL.eval() >= 0.
@pytest.mark.parametrize('dists', [
{'prior_W': norm_prior(DIM, 1.), 'prior_b': norm_prior((D,), 1.)},
{'post_W': norm_prior(DIM, 1.), 'post_b': norm_prior((D,), 1.)},
{'prior_W': norm_prior(DIM, 1.), 'post_W': norm_prior(DIM, 1.)},
{'prior_W': norm_prior(DIM, 1.), 'post_W': gaus_posterior(DIM, 1.)},
{'prior_W': gaus_posterior(DIM, 1.), 'post_W': gaus_posterior(DIM, 1.)},
])
def test_dense_distribution(dists, make_data):
"""Test initialising dense variational layers with distributions."""
x, _, _ = make_data
S = 3
x_, X_ = _make_placeholders(x, S)
N = x.shape[0]
Phi, KL = ab.DenseVariational(output_dim=D, **dists)(X_)
tc = tf.test.TestCase()
with tc.test_session():
tf.global_variables_initializer().run()
P = Phi.eval(feed_dict={x_: x})