def test_bad_kernel_approximation(self, initializer, scale, exact_kernel_fn):
"""Approximation is bad when output dimension is small."""
# Two distinct inputs.
x = tf.constant([[1.0, -1.0, 0.5]])
y = tf.constant([[-1.0, 1.0, 1.0]])
small_output_dim = 10
tf.compat.v1.set_random_seed(1234)
# Initialize layer.
rff_layer = kernel_layers.RandomFourierFeatures(
output_dim=small_output_dim,
kernel_initializer=initializer,
scale=scale,
name='random_fourier_features')
# Apply layer to both inputs.
output_x = math.sqrt(2.0 / small_output_dim) * rff_layer(x)
output_y = math.sqrt(2.0 / small_output_dim) * rff_layer(y)
# The inner products of the outputs (on inputs x and y) approximates the
# real value of the RBF kernel but poorly since the output dimension of the
# layer is small.
exact_kernel_value = exact_kernel_fn(x, y)
approx_kernel_value = kernelized_utils.inner_product(output_x, output_y)
abs_error = tf.abs(exact_kernel_value - approx_kernel_value)
if not tf.executing_eagerly():
with self.cached_session() as sess:
keras_backend._initialize_variables(sess)
abs_error_eval = sess.run([abs_error])
self.assertGreater(abs_error_eval[0][0], 0.01)
self.assertLess(abs_error_eval[0][0], 0.5)
else:
self.assertGreater(abs_error, 0.01)
self.assertLess(abs_error, 0.5)
def _assert_all_close(self, expected, actual):
if not tf.executing_eagerly():
with self.cached_session() as sess:
backend._initialize_variables(sess)
self.assertAllClose(expected, actual)
else:
self.assertAllClose(expected, actual)
def init_restore_or_wait_for_variables():
"""Initialize or restore variables or wait for variables to be initialized."""
session = backend._get_session() # pylint: disable=protected-access
if not multi_worker_util.has_worker_context(
) or multi_worker_util.should_load_checkpoint():
# TODO(yuefengz): if checkpoints exist, restore from checkpoint.
backend._initialize_variables(session) # pylint: disable=protected-access
else:
_wait_for_variable_initialization(session)
def _build(self, shape):
self._shape = tf.TensorShape(shape)
self._build_input_shape = self._shape
# Create new state variables
self._total = self.add_weight(
name='total', shape=shape, initializer='zeros')
self._count = self.add_weight(
name='count', shape=shape, initializer='zeros')
with tf.init_scope():
if not tf.executing_eagerly():
backend._initialize_variables(backend._get_session()) # pylint: disable=protected-access
self._built = True
def _build(self, shape):
self._shape = tf.TensorShape(shape)
self._build_input_shape = self._shape
# Create new state variables
self._total = self.add_weight(
name="total", shape=shape, initializer="zeros"
)
self._count = self.add_weight(
name="count", shape=shape, initializer="zeros"
)
with tf.init_scope():
if not tf.executing_eagerly():
backend._initialize_variables(backend._get_session())
self._built = True
def init_restore_or_wait_for_variables():
"""Initialize or restore variables or wait for variables to be initialized."""
backend._initialize_variables(backend._get_session()) # pylint: disable=protected-access
def init_restore_or_wait_for_variables():
"""Initialize or restore variables or wait for variables to be
initialized."""
backend._initialize_variables(backend._get_session())
