def reparam(rdft):
irdft_matrix = spectral_ops.irdft_matrix(var_shape[:-2],
dtype=var_dtype)
if not self.dc:
irdft_matrix = irdft_matrix[:, 1:]
var = tf.linalg.matmul(irdft_matrix, rdft)
var = tf.reshape(var, var_shape)
return var
def test_irdft3_matrix(self):
for shape in [(3, 4, 2), (6, 3, 1)]:
size = shape[0] * shape[1] * shape[2]
matrix = spectral_ops.irdft_matrix(shape)
# Test that the matrix is orthonormal.
result = tf.matmul(matrix, tf.transpose(matrix))
with self.test_session() as sess:
result, = sess.run([result])
self.assertAllClose(result, np.identity(size))
def test_irdft3_matrix(self):
for shape in [(3, 4, 2), (6, 3, 1)]:
size = shape[0] * shape[1] * shape[2]
matrix = spectral_ops.irdft_matrix(shape)
# Test that the matrix is orthonormal.
result = tf.matmul(matrix, tf.transpose(matrix))
with self.test_session() as sess:
result, = sess.run([result])
self.assertAllClose(result, np.identity(size))
def __call__(self,
getter,
name,
shape,
dtype,
initializer,
regularizer=None):
if all(s == 1 for s in shape[:-2]):
return getter(name=name,
shape=shape,
dtype=dtype,
initializer=initializer,
regularizer=regularizer)
var_shape = shape
var_dtype = dtype
size = var_shape[0]
for s in var_shape[1:-2]:
size *= s
irdft_matrix = spectral_ops.irdft_matrix(var_shape[:-2],
dtype=var_dtype)
if self.dc:
rdft_shape = (size, var_shape[-2] * var_shape[-1])
else:
irdft_matrix = irdft_matrix[:, 1:]
rdft_shape = (size - 1, var_shape[-2] * var_shape[-1])
rdft_dtype = var_dtype
rdft_name = name + "_rdft"
def rdft_initializer(shape, dtype=None, partition_info=None):
assert tuple(shape) == rdft_shape, shape
assert dtype == rdft_dtype, dtype
init = initializer(var_shape,
dtype=var_dtype,
partition_info=partition_info)
init = tf.reshape(init, (-1, rdft_shape[-1]))
init = tf.linalg.matmul(irdft_matrix, init, transpose_a=True)
return init
def reparam(rdft):
var = tf.linalg.matmul(irdft_matrix, rdft)
var = tf.reshape(var, var_shape)
return var
if regularizer is not None:
regularizer = lambda rdft: regularizer(reparam(rdft))
rdft = getter(name=rdft_name,
shape=rdft_shape,
dtype=rdft_dtype,
initializer=rdft_initializer,
regularizer=regularizer)
return reparam(rdft)
def rdft_initializer(shape, dtype=None, partition_info=None):
"""Initializer wrapper."""
del partition_info # Ignored for TF 1/2 compatibility.
assert tuple(shape) == rdft_shape, shape
assert dtype == rdft_dtype, dtype
init = initializer(var_shape, dtype=var_dtype)
init = tf.reshape(init, (-1, rdft_shape[-1]))
irdft_matrix = spectral_ops.irdft_matrix(var_shape[:-2],
dtype=var_dtype)
if not self.dc:
irdft_matrix = irdft_matrix[:, 1:]
init = tf.linalg.matmul(irdft_matrix, init, transpose_a=True)
return init
def __init__(self,
initial_value,
name=None,
dc=True,
shape=None,
dtype=None):
"""Initializer.
Args:
initial_value: `tf.Tensor` or `None`. The initial value of the kernel. If
not provided, its `shape` must be given, and the initial value of the
parameter will be undefined.
name: String. The name of the kernel.
dc: Boolean. If `False`, the DC component of the kernel RDFTs is not
represented, forcing the filters to be highpass. Defaults to `True`.
shape: `tf.TensorShape` or compatible. Ignored unless `initial_value is
None`.
dtype: `tf.dtypes.DType` or compatible. DType of this parameter. If not
given, inferred from `initial_value`.
"""
super().__init__(name=name)
self._dc = bool(dc)
if initial_value is None:
if shape is None:
raise ValueError(
"If initial_value is None, shape must be specified.")
initial_value = tf.zeros(shape, dtype=dtype)
else:
initial_value = tf.convert_to_tensor(initial_value, dtype=dtype)
self._shape = initial_value.shape
self._matrix = spectral_ops.irdft_matrix(self.shape[:-2],
dtype=initial_value.dtype)
if not self.dc:
self._matrix = self._matrix[:, 1:]
initial_value = tf.reshape(initial_value,
(-1, self.shape[-2] * self.shape[-1]))
initial_value = tf.linalg.matmul(self._matrix,
initial_value,
transpose_a=True)
if name is not None:
name = f"{name}_rdft"
self.rdft = tf.Variable(initial_value, name=name)
def __call__(self, getter, name, shape, dtype, initializer, regularizer=None):
if all(s == 1 for s in shape[:-2]):
return getter(name=name, shape=shape, dtype=dtype,
initializer=initializer, regularizer=regularizer)
var_shape = shape
var_dtype = dtype
size = var_shape[0]
for s in var_shape[1:-2]:
size *= s
irdft_matrix = spectral_ops.irdft_matrix(var_shape[:-2], dtype=var_dtype)
if self.dc:
rdft_shape = (size, var_shape[-2] * var_shape[-1])
else:
irdft_matrix = irdft_matrix[:, 1:]
rdft_shape = (size - 1, var_shape[-2] * var_shape[-1])
rdft_dtype = var_dtype
rdft_name = name + "_rdft"
def rdft_initializer(shape, dtype=None, partition_info=None):
assert tuple(shape) == rdft_shape, shape
assert dtype == rdft_dtype, dtype
init = initializer(
var_shape, dtype=var_dtype, partition_info=partition_info)
init = tf.reshape(init, (-1, rdft_shape[-1]))
init = tf.linalg.matmul(irdft_matrix, init, transpose_a=True)
return init
def reparam(rdft):
var = tf.linalg.matmul(irdft_matrix, rdft)
var = tf.reshape(var, var_shape)
return var
if regularizer is not None:
regularizer = lambda rdft: regularizer(reparam(rdft))
rdft = getter(
name=rdft_name, shape=rdft_shape, dtype=rdft_dtype,
initializer=rdft_initializer, regularizer=regularizer)
return reparam(rdft)
def test_irdft_matrix_is_orthonormal(self, *shape): matrix = spectral_ops.irdft_matrix(shape) result = tf.matmul(matrix, tf.transpose(matrix)) self.assertAllClose(result, tf.eye(tf.TensorShape(shape).num_elements()))
