def test_variable_reuse(self):
"""Test that variable scopes work and inference on a real-ish case."""
if tf.executing_eagerly():
# Variable reuse doesn't work in eager.
return
tensor1_ref = tf.zeros([6, 5, 7, 3, 3])
tensor1_examples = tf.zeros([4, 5, 7, 3, 3])
tensor2_ref = tf.zeros([4, 2, 3])
tensor2_examples = tf.zeros([2, 2, 3])
vbn1 = vbn_lib.VBN(tensor1_ref, name='vbn1')
vbn2 = vbn_lib.VBN(tensor2_ref, name='vbn2')
# Fetch reference and examples after virtual batch normalization. Also
# fetch in variable reuse case.
to_fetch = []
to_fetch.append(vbn1.reference_batch_normalization())
to_fetch.append(vbn2.reference_batch_normalization())
to_fetch.append(vbn1(tensor1_examples))
to_fetch.append(vbn2(tensor2_examples))
tf.compat.v1.get_variable_scope().reuse_variables()
to_fetch.append(vbn1.reference_batch_normalization())
to_fetch.append(vbn2.reference_batch_normalization())
to_fetch.append(vbn1(tensor1_examples))
to_fetch.append(vbn2(tensor2_examples))
self.assertLen(contrib.get_variables(), 4)
with self.cached_session() as sess:
sess.run(tf.compat.v1.global_variables_initializer())
sess.run(to_fetch)
def test_virtual_statistics(self):
"""Check that `_virtual_statistics` gives same result as `nn.moments`."""
tf.compat.v1.set_random_seed(1234)
batch_axis = 0
partial_batch = tf.random.normal([4, 5, 7, 3])
single_example = tf.random.normal([1, 5, 7, 3])
full_batch = tf.concat([partial_batch, single_example], axis=0)
for reduction_axis in range(1, 4):
# Get `nn.moments` on the full batch.
reduction_axes = list(range(4))
del reduction_axes[reduction_axis]
mom_mean, mom_variance = tf.nn.moments(x=full_batch, axes=reduction_axes)
# Get virtual batch statistics.
vb_reduction_axes = list(range(4))
del vb_reduction_axes[reduction_axis]
del vb_reduction_axes[batch_axis]
vbn = vbn_lib.VBN(partial_batch, reduction_axis)
vb_mean, mean_sq = vbn._virtual_statistics(single_example,
vb_reduction_axes)
vb_variance = mean_sq - tf.square(vb_mean)
# Remove singleton batch dim for easy comparisons.
vb_mean = tf.squeeze(vb_mean, batch_axis)
vb_variance = tf.squeeze(vb_variance, batch_axis)
with self.cached_session() as sess:
vb_mean_np, vb_var_np, mom_mean_np, mom_var_np = sess.run(
[vb_mean, vb_variance, mom_mean, mom_variance])
self.assertAllClose(mom_mean_np, vb_mean_np)
self.assertAllClose(mom_var_np, vb_var_np)
def test_minibatch_independent(self):
"""Test that virtual batch normalized examples are independent.
Unlike batch normalization, virtual batch normalization has the property
that the virtual batch normalized value of an example is independent of the
other examples in the minibatch. In this test, we verify this property.
"""
tf.compat.v1.set_random_seed(1234)
# These can be random, but must be the same for all session calls.
reference_batch = tf.constant(
np.random.normal(size=[4, 7, 3]), dtype=tf.float32)
fixed_example = tf.constant(np.random.normal(size=[7, 3]), dtype=tf.float32)
# Get the VBN object and the virtual batch normalized value for
# `fixed_example`.
vbn = vbn_lib.VBN(reference_batch)
vbn_fixed_example = tf.squeeze(vbn(tf.expand_dims(fixed_example, 0)), 0)
with self.cached_session() as sess:
sess.run(tf.compat.v1.global_variables_initializer())
vbn_fixed_example_np = sess.run(vbn_fixed_example)
# Check that the value is the same for different minibatches, and different
# sized minibatches.
for minibatch_size in range(1, 6):
examples = [tf.random.normal([7, 3]) for _ in range(minibatch_size)]
minibatch = tf.stack([fixed_example] + examples)
vbn_minibatch = vbn(minibatch)
cur_vbn_fixed_example = vbn_minibatch[0, ...]
with self.cached_session() as sess:
sess.run(tf.compat.v1.global_variables_initializer())
cur_vbn_fixed_example_np = sess.run(cur_vbn_fixed_example)
self.assertAllClose(vbn_fixed_example_np, cur_vbn_fixed_example_np)
def test_variable_reuse_fail(self):
"""Test that incorrect usage causes variable fail."""
if tf.executing_eagerly():
# Variable reuse doesn't work in eager.
return
tensor1_ref = tf.zeros([6, 5, 7, 3, 3])
with tf.compat.v1.variable_scope('dummy_scope', reuse=True):
with self.assertRaisesRegexp(
ValueError, 'does not exist, or was not created with '
'tf.get_variable()'):
vbn_lib.VBN(tensor1_ref)
def test_same_as_batchnorm(self):
"""Check that batch norm on set X is the same as ref of X / y on `y`."""
tf.compat.v1.set_random_seed(1234)
num_examples = 4
examples = [tf.random.normal([5, 7, 3]) for _ in range(num_examples)]
# Get the result of the opensource batch normalization.
batch_normalized = tf.compat.v1.layers.batch_normalization(
tf.stack(examples), training=True)
for i in range(num_examples):
examples_except_i = tf.stack(examples[:i] + examples[i + 1:])
# Get the result of VBN's batch normalization.
vbn = vbn_lib.VBN(examples_except_i)
vb_normed = tf.squeeze(vbn(tf.expand_dims(examples[i], [0])), [0])
with self.cached_session() as sess:
sess.run(tf.compat.v1.global_variables_initializer())
bn_np, vb_np = sess.run([batch_normalized, vb_normed])
self.assertAllClose(bn_np[i, ...], vb_np)
def test_reference_batch_normalization(self):
"""Check that batch norm from VBN agrees with opensource implementation."""
tf.compat.v1.set_random_seed(1234)
batch = tf.random.normal([6, 5, 7, 3, 3])
for axis in range(5):
# Get `layers` batchnorm result.
bn_normalized = tf.compat.v1.layers.batch_normalization(
batch, axis, training=True)
# Get VBN's batch normalization on reference batch.
batch_axis = 0 if axis is not 0 else 1 # axis and batch_axis can't same
vbn = vbn_lib.VBN(batch, axis, batch_axis=batch_axis)
vbn_normalized = vbn.reference_batch_normalization()
with self.cached_session() as sess:
sess.run(tf.compat.v1.global_variables_initializer())
bn_normalized_np, vbn_normalized_np = sess.run(
[bn_normalized, vbn_normalized])
self.assertAllClose(bn_normalized_np, vbn_normalized_np)
def test_invalid_input(self):
# Reference batch has unknown dimensions.
if not tf.executing_eagerly():
with self.assertRaisesRegexp(ValueError,
'`reference_batch` has unknown dimensions.'):
vbn_lib.VBN(tf.compat.v1.placeholder(tf.float32), name='vbn1')
# Axis too negative.
with self.assertRaisesRegexp(ValueError,
'Value of `axis` argument .* is out of range'):
vbn_lib.VBN(tf.zeros([1, 2]), axis=-3, name='vbn2')
# Axis too large.
with self.assertRaisesRegexp(ValueError,
'Value of `axis` argument .* is out of range'):
vbn_lib.VBN(tf.zeros([1, 2]), axis=2, name='vbn3')
# Batch axis too negative.
with self.assertRaisesRegexp(ValueError,
'Value of `axis` argument .* is out of range'):
vbn_lib.VBN(tf.zeros([1, 2]), name='vbn4', batch_axis=-3)
# Batch axis too large.
with self.assertRaisesRegexp(ValueError,
'Value of `axis` argument .* is out of range'):
vbn_lib.VBN(tf.zeros([1, 2]), name='vbn5', batch_axis=2)
# Axis and batch axis are the same.
with self.assertRaisesRegexp(ValueError,
'`axis` and `batch_axis` cannot be the same.'):
vbn_lib.VBN(tf.zeros([1, 2]), axis=1, name='vbn6', batch_axis=1)
# Reference Tensor and example Tensor have incompatible shapes.
tensor_ref = tf.zeros([5, 2, 3])
tensor_examples = tf.zeros([3, 2, 3])
vbn = vbn_lib.VBN(tensor_ref, name='vbn7', batch_axis=1)
with self.assertRaisesRegexp(ValueError, 'Shapes .* are incompatible'):
vbn(tensor_examples)
def test_no_broadcast_needed(self): """When `axis` and `batch_axis` are at the end, no broadcast is needed.""" reference_batch = tf.zeros([5, 3, 16, 9, 15]) minibatch = tf.zeros([5, 3, 16, 3, 15]) vbn = vbn_lib.VBN(reference_batch, axis=-1, batch_axis=-2) vbn(minibatch)
def test_syntax(self): reference_batch = tf.zeros([5, 3, 16, 9, 15]) vbn = vbn_lib.VBN(reference_batch, batch_axis=1) vbn(tf.ones([5, 7, 16, 9, 15]))
