Python tensorflow.python.ops.array_ops.matrix_diag() Examples

The following are 30 code examples of tensorflow.python.ops.array_ops.matrix_diag(). You can vote up the ones you like or vote down the ones you don't like, and go to the original project or source file by following the links above each example. You may also want to check out all available functions/classes of the module tensorflow.python.ops.array_ops , or try the search function .
Example #1
Source File: linalg_grad.py    From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License 5 votes vote down vote up 
def _SelfAdjointEigV2Grad(op, grad_e, grad_v):
  """Gradient for SelfAdjointEigV2."""
  e = op.outputs[0]
  compute_v = op.get_attr("compute_v")
  # a = op.inputs[0], which satisfies
  # a[...,:,:] * v[...,:,i] = e[...,i] * v[...,i]
  with ops.control_dependencies([grad_e, grad_v]):
    if compute_v:
      v = op.outputs[1]
      # Construct the matrix f(i,j) = (i != j ? 1 / (e_i - e_j) : 0).
      # Notice that because of the term involving f, the gradient becomes
      # infinite (or NaN in practice) when eigenvalues are not unique.
      # Mathematically this should not be surprising, since for (k-fold)
      # degenerate eigenvalues, the corresponding eigenvectors are only defined
      # up to arbitrary rotation in a (k-dimensional) subspace.
      f = array_ops.matrix_set_diag(
          math_ops.reciprocal(
              array_ops.expand_dims(e, -2) - array_ops.expand_dims(e, -1)),
          array_ops.zeros_like(e))
      grad_a = math_ops.matmul(
          v,
          math_ops.matmul(
              array_ops.matrix_diag(grad_e) +
              f * math_ops.matmul(v, grad_v, adjoint_a=True),
              v,
              adjoint_b=True))
    else:
      _, v = linalg_ops.self_adjoint_eig(op.inputs[0])
      grad_a = math_ops.matmul(v,
                               math_ops.matmul(
                                   array_ops.matrix_diag(grad_e),
                                   v,
                                   adjoint_b=True))
    # The forward op only depends on the lower triangular part of a, so here we
    # symmetrize and take the lower triangle
    grad_a = array_ops.matrix_band_part(
        grad_a + math_ops.conj(array_ops.matrix_transpose(grad_a)), -1, 0)
    grad_a = array_ops.matrix_set_diag(grad_a,
                                       0.5 * array_ops.matrix_diag_part(grad_a))
    return grad_a
Example #2
Source File: array_grad.py    From lambda-packs with MIT License 5 votes vote down vote up 
def _MatrixDiagPartGrad(op, grad):
  matrix_shape = op.inputs[0].get_shape()[-2:]
  if matrix_shape.is_fully_defined() and matrix_shape[0] == matrix_shape[1]:
    return array_ops.matrix_diag(grad)
  else:
    return array_ops.matrix_set_diag(array_ops.zeros_like(op.inputs[0]), grad)
Example #3
Source File: array_grad.py    From deep_image_model with Apache License 2.0 5 votes vote down vote up 
def _MatrixDiagPartGrad(op, grad):
  matrix_shape = op.inputs[0].get_shape()[-2:]
  if matrix_shape.is_fully_defined() and matrix_shape[0] == matrix_shape[1]:
    return array_ops.matrix_diag(grad)
  else:
    return array_ops.matrix_set_diag(array_ops.zeros_like(op.inputs[0]), grad)
Example #4
Source File: linalg_grad.py    From deep_image_model with Apache License 2.0 5 votes vote down vote up 
def _SelfAdjointEigV2Grad(op, grad_e, grad_v):
  """Gradient for SelfAdjointEigV2."""
  e = op.outputs[0]
  v = op.outputs[1]
  # a = op.inputs[0], which satisfies
  # a[...,:,:] * v[...,:,i] = e[...,i] * v[...,i]
  with ops.control_dependencies([grad_e.op, grad_v.op]):
    if grad_v is not None:
      # Construct the matrix f(i,j) = (i != j ? 1 / (e_i - e_j) : 0).
      # Notice that because of the term involving f, the gradient becomes
      # infinite (or NaN in practice) when eigenvalues are not unique.
      # Mathematically this should not be surprising, since for (k-fold)
      # degenerate eigenvalues, the corresponding eigenvectors are only defined
      # up to arbitrary rotation in a (k-dimensional) subspace.
      f = array_ops.matrix_set_diag(
          math_ops.inv(
              array_ops.expand_dims(e, -2) - array_ops.expand_dims(e, -1)),
          array_ops.zeros_like(e))
      grad_a = math_ops.batch_matmul(
          v,
          math_ops.batch_matmul(
              array_ops.matrix_diag(grad_e) + f * math_ops.batch_matmul(
                  v, grad_v, adj_x=True),
              v,
              adj_y=True))
    else:
      grad_a = math_ops.batch_matmul(
          v,
          math_ops.batch_matmul(
              array_ops.matrix_diag(grad_e), v, adj_y=True))
    # The forward op only depends on the lower triangular part of a, so here we
    # symmetrize and take the lower triangle
    grad_a = array_ops.matrix_band_part(
        grad_a + array_ops.matrix_transpose(grad_a), -1, 0)
    grad_a = array_ops.matrix_set_diag(grad_a,
                                       0.5 * array_ops.matrix_diag_part(grad_a))
    return grad_a
Example #5
Source File: operator_pd_identity.py    From deep_image_model with Apache License 2.0 5 votes vote down vote up 
def _to_dense(self):
    diag = array_ops.ones(self.vector_shape(), dtype=self.dtype)
    dense = array_ops.matrix_diag(diag)
    dense.set_shape(self.get_shape())
    return dense
Example #6
Source File: operator_pd_diag.py    From deep_image_model with Apache License 2.0 5 votes vote down vote up 
def _to_dense(self):
    return array_ops.matrix_diag(self._diag)
Example #7
Source File: operator_pd_diag.py    From deep_image_model with Apache License 2.0 5 votes vote down vote up 
def _to_dense(self):
    return array_ops.matrix_diag(math_ops.square(self._diag))
Example #8
Source File: operator_pd_diag.py    From deep_image_model with Apache License 2.0 5 votes vote down vote up 
def _sqrt_to_dense(self):
    return array_ops.matrix_diag(self._diag)
Example #9
Source File: array_grad.py    From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License 5 votes vote down vote up 
def _MatrixDiagPartGrad(op, grad):
  matrix_shape = op.inputs[0].get_shape()[-2:]
  if matrix_shape.is_fully_defined() and matrix_shape[0] == matrix_shape[1]:
    return array_ops.matrix_diag(grad)
  else:
    return array_ops.matrix_set_diag(array_ops.zeros_like(op.inputs[0]), grad)
Example #10
Source File: linear_operator_diag.py    From auto-alt-text-lambda-api with MIT License 5 votes vote down vote up 
def _to_dense(self):
    return array_ops.matrix_diag(self._diag)
Example #11
Source File: array_grad.py    From keras-lambda with MIT License 5 votes vote down vote up 
def _MatrixDiagPartGrad(op, grad):
  matrix_shape = op.inputs[0].get_shape()[-2:]
  if matrix_shape.is_fully_defined() and matrix_shape[0] == matrix_shape[1]:
    return array_ops.matrix_diag(grad)
  else:
    return array_ops.matrix_set_diag(array_ops.zeros_like(op.inputs[0]), grad)
Example #12
Source File: linalg_grad.py    From keras-lambda with MIT License 5 votes vote down vote up 
def _SelfAdjointEigV2Grad(op, grad_e, grad_v):
  """Gradient for SelfAdjointEigV2."""
  e = op.outputs[0]
  v = op.outputs[1]
  # a = op.inputs[0], which satisfies
  # a[...,:,:] * v[...,:,i] = e[...,i] * v[...,i]
  with ops.control_dependencies([grad_e.op, grad_v.op]):
    if grad_v is not None:
      # Construct the matrix f(i,j) = (i != j ? 1 / (e_i - e_j) : 0).
      # Notice that because of the term involving f, the gradient becomes
      # infinite (or NaN in practice) when eigenvalues are not unique.
      # Mathematically this should not be surprising, since for (k-fold)
      # degenerate eigenvalues, the corresponding eigenvectors are only defined
      # up to arbitrary rotation in a (k-dimensional) subspace.
      f = array_ops.matrix_set_diag(
          math_ops.reciprocal(
              array_ops.expand_dims(e, -2) - array_ops.expand_dims(e, -1)),
          array_ops.zeros_like(e))
      grad_a = math_ops.matmul(
          v,
          math_ops.matmul(
              array_ops.matrix_diag(grad_e) + f * math_ops.matmul(
                  v, grad_v, adjoint_a=True),
              v,
              adjoint_b=True))
    else:
      grad_a = math_ops.matmul(
          v, math_ops.matmul(
              array_ops.matrix_diag(grad_e), v, adjoint_b=True))
    # The forward op only depends on the lower triangular part of a, so here we
    # symmetrize and take the lower triangle
    grad_a = array_ops.matrix_band_part(
        grad_a + array_ops.matrix_transpose(grad_a), -1, 0)
    grad_a = array_ops.matrix_set_diag(grad_a,
                                       0.5 * array_ops.matrix_diag_part(grad_a))
    return grad_a
Example #13
Source File: operator_pd_identity.py    From keras-lambda with MIT License 5 votes vote down vote up 
def _to_dense(self):
    diag = array_ops.ones(self.vector_shape(), dtype=self.dtype)
    dense = array_ops.matrix_diag(diag)
    dense.set_shape(self.get_shape())
    return self._scale * dense
Example #14
Source File: operator_pd_identity.py    From keras-lambda with MIT License 5 votes vote down vote up 
def _sqrt_to_dense(self):
    diag = array_ops.ones(self.vector_shape(), dtype=self.dtype)
    dense = array_ops.matrix_diag(diag)
    dense.set_shape(self.get_shape())
    return math_ops.sqrt(self._scale) * dense
Example #15
Source File: operator_pd_diag.py    From keras-lambda with MIT License 5 votes vote down vote up 
def _sqrt_to_dense(self):
    return array_ops.matrix_diag(math_ops.sqrt(self._diag))
Example #16
Source File: operator_pd_diag.py    From keras-lambda with MIT License 5 votes vote down vote up 
def _to_dense(self):
    return array_ops.matrix_diag(math_ops.square(self._diag))
Example #17
Source File: operator_pd_diag.py    From keras-lambda with MIT License 5 votes vote down vote up 
def _sqrt_to_dense(self):
    return array_ops.matrix_diag(self._diag)
Example #18
Source File: linear_operator_diag.py    From keras-lambda with MIT License 5 votes vote down vote up 
def _to_dense(self):
    return array_ops.matrix_diag(self._diag)
Example #19
Source File: operator_pd_diag.py    From auto-alt-text-lambda-api with MIT License 5 votes vote down vote up 
def _to_dense(self):
    return array_ops.matrix_diag(math_ops.square(self._diag))
Example #20
Source File: linalg_grad.py    From lambda-packs with MIT License 5 votes vote down vote up 
def _SelfAdjointEigV2Grad(op, grad_e, grad_v):
  """Gradient for SelfAdjointEigV2."""
  e = op.outputs[0]
  v = op.outputs[1]
  # a = op.inputs[0], which satisfies
  # a[...,:,:] * v[...,:,i] = e[...,i] * v[...,i]
  with ops.control_dependencies([grad_e.op, grad_v.op]):
    if grad_v is not None:
      # Construct the matrix f(i,j) = (i != j ? 1 / (e_i - e_j) : 0).
      # Notice that because of the term involving f, the gradient becomes
      # infinite (or NaN in practice) when eigenvalues are not unique.
      # Mathematically this should not be surprising, since for (k-fold)
      # degenerate eigenvalues, the corresponding eigenvectors are only defined
      # up to arbitrary rotation in a (k-dimensional) subspace.
      f = array_ops.matrix_set_diag(
          math_ops.reciprocal(
              array_ops.expand_dims(e, -2) - array_ops.expand_dims(e, -1)),
          array_ops.zeros_like(e))
      grad_a = math_ops.matmul(
          v,
          math_ops.matmul(
              array_ops.matrix_diag(grad_e) + f * math_ops.matmul(
                  v, grad_v, adjoint_a=True),
              v,
              adjoint_b=True))
    else:
      grad_a = math_ops.matmul(
          v, math_ops.matmul(
              array_ops.matrix_diag(grad_e), v, adjoint_b=True))
    # The forward op only depends on the lower triangular part of a, so here we
    # symmetrize and take the lower triangle
    grad_a = array_ops.matrix_band_part(
        grad_a + array_ops.matrix_transpose(grad_a), -1, 0)
    grad_a = array_ops.matrix_set_diag(grad_a,
                                       0.5 * array_ops.matrix_diag_part(grad_a))
    return grad_a
Example #21
Source File: operator_pd_identity.py    From lambda-packs with MIT License 5 votes vote down vote up 
def _to_dense(self):
    diag = array_ops.ones(self.vector_shape(), dtype=self.dtype)
    dense = array_ops.matrix_diag(diag)
    dense.set_shape(self.get_shape())
    return self._scale * dense
Example #22
Source File: operator_pd_identity.py    From lambda-packs with MIT License 5 votes vote down vote up 
def _sqrt_to_dense(self):
    diag = array_ops.ones(self.vector_shape(), dtype=self.dtype)
    dense = array_ops.matrix_diag(diag)
    dense.set_shape(self.get_shape())
    return math_ops.sqrt(self._scale) * dense
Example #23
Source File: mvn_linear_operator.py    From lambda-packs with MIT License 5 votes vote down vote up 
def _covariance(self):
    if distribution_util.is_diagonal_scale(self.scale):
      return array_ops.matrix_diag(math_ops.square(self.scale.diag_part()))
    else:
      return self.scale.matmul(self.scale.to_dense(), adjoint_arg=True)
Example #24
Source File: operator_pd_diag.py    From lambda-packs with MIT License 5 votes vote down vote up 
def _to_dense(self):
    return array_ops.matrix_diag(self._diag)
Example #25
Source File: operator_pd_diag.py    From lambda-packs with MIT License 5 votes vote down vote up 
def _sqrt_to_dense(self):
    return array_ops.matrix_diag(math_ops.sqrt(self._diag))
Example #26
Source File: operator_pd_diag.py    From lambda-packs with MIT License 5 votes vote down vote up 
def _to_dense(self):
    return array_ops.matrix_diag(math_ops.square(self._diag))
Example #27
Source File: operator_pd_diag.py    From lambda-packs with MIT License 5 votes vote down vote up 
def _sqrt_to_dense(self):
    return array_ops.matrix_diag(self._diag)
Example #28
Source File: array_grad.py    From auto-alt-text-lambda-api with MIT License 5 votes vote down vote up 
def _MatrixDiagPartGrad(op, grad):
  matrix_shape = op.inputs[0].get_shape()[-2:]
  if matrix_shape.is_fully_defined() and matrix_shape[0] == matrix_shape[1]:
    return array_ops.matrix_diag(grad)
  else:
    return array_ops.matrix_set_diag(array_ops.zeros_like(op.inputs[0]), grad)
Example #29
Source File: linalg_grad.py    From auto-alt-text-lambda-api with MIT License 5 votes vote down vote up 
def _SelfAdjointEigV2Grad(op, grad_e, grad_v):
  """Gradient for SelfAdjointEigV2."""
  e = op.outputs[0]
  v = op.outputs[1]
  # a = op.inputs[0], which satisfies
  # a[...,:,:] * v[...,:,i] = e[...,i] * v[...,i]
  with ops.control_dependencies([grad_e.op, grad_v.op]):
    if grad_v is not None:
      # Construct the matrix f(i,j) = (i != j ? 1 / (e_i - e_j) : 0).
      # Notice that because of the term involving f, the gradient becomes
      # infinite (or NaN in practice) when eigenvalues are not unique.
      # Mathematically this should not be surprising, since for (k-fold)
      # degenerate eigenvalues, the corresponding eigenvectors are only defined
      # up to arbitrary rotation in a (k-dimensional) subspace.
      f = array_ops.matrix_set_diag(
          math_ops.reciprocal(
              array_ops.expand_dims(e, -2) - array_ops.expand_dims(e, -1)),
          array_ops.zeros_like(e))
      grad_a = math_ops.matmul(
          v,
          math_ops.matmul(
              array_ops.matrix_diag(grad_e) + f * math_ops.matmul(
                  v, grad_v, adjoint_a=True),
              v,
              adjoint_b=True))
    else:
      grad_a = math_ops.matmul(
          v, math_ops.matmul(
              array_ops.matrix_diag(grad_e), v, adjoint_b=True))
    # The forward op only depends on the lower triangular part of a, so here we
    # symmetrize and take the lower triangle
    grad_a = array_ops.matrix_band_part(
        grad_a + array_ops.matrix_transpose(grad_a), -1, 0)
    grad_a = array_ops.matrix_set_diag(grad_a,
                                       0.5 * array_ops.matrix_diag_part(grad_a))
    return grad_a
Example #30
Source File: operator_pd_identity.py    From auto-alt-text-lambda-api with MIT License 5 votes vote down vote up 
def _to_dense(self):
    diag = array_ops.ones(self.vector_shape(), dtype=self.dtype)
    dense = array_ops.matrix_diag(diag)
    dense.set_shape(self.get_shape())
    return self._scale * dense