Python array_ops.matrix_band_part方法代码示例

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def _MatrixTriangularSolveGrad(op, grad):
  """Gradient for MatrixTriangularSolve."""
  a = op.inputs[0]
  adjoint_a = op.get_attr("adjoint")
  lower_a = op.get_attr("lower")
  c = op.outputs[0]
  grad_b = linalg_ops.matrix_triangular_solve(
      a, grad, lower=lower_a, adjoint=not adjoint_a)
  if adjoint_a:
    grad_a = -math_ops.matmul(c, grad_b, adjoint_b=True)
  else:
    grad_a = -math_ops.matmul(grad_b, c, adjoint_b=True)
  if lower_a:
    grad_a = array_ops.matrix_band_part(grad_a, -1, 0)
  else:
    grad_a = array_ops.matrix_band_part(grad_a, 0, -1)
  return (grad_a, grad_b) 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def _MatrixTriangularSolveGrad(op, grad):
  """Gradient for MatrixTriangularSolve."""
  a = op.inputs[0]
  adjoint_a = op.get_attr("adjoint")
  lower_a = op.get_attr("lower")
  c = op.outputs[0]
  grad_b = linalg_ops.matrix_triangular_solve(
      a, grad, lower=lower_a, adjoint=not adjoint_a)
  if adjoint_a:
    grad_a = -math_ops.batch_matmul(c, grad_b, adj_y=True)
  else:
    grad_a = -math_ops.batch_matmul(grad_b, c, adj_y=True)
  if lower_a:
    grad_a = array_ops.matrix_band_part(grad_a, -1, 0)
  else:
    grad_a = array_ops.matrix_band_part(grad_a, 0, -1)
  return (grad_a, grad_b) 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def _CholeskyGrad(op, grad):
  """Gradient for Cholesky."""

  # Gradient is l^{-H} @ ((l^{H} @ grad) * (tril(ones)-1/2*eye)) @ l^{-1}
  l = op.outputs[0]
  num_rows = array_ops.shape(l)[-1]
  batch_shape = array_ops.shape(l)[:-2]
  l_inverse = linalg_ops.matrix_triangular_solve(l,
                                                 linalg_ops.eye(
                                                     num_rows,
                                                     batch_shape=batch_shape,
                                                     dtype=l.dtype))

  middle = math_ops.matmul(l, grad, adjoint_a=True)
  middle = array_ops.matrix_set_diag(middle,
                                     0.5 * array_ops.matrix_diag_part(middle))
  middle = array_ops.matrix_band_part(middle, -1, 0)

  grad_a = math_ops.matmul(
      math_ops.matmul(l_inverse, middle, adjoint_a=True), l_inverse)

  grad_a += math_ops.conj(array_ops.matrix_transpose(grad_a))
  return grad_a * 0.5 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def _MatrixBandPartGrad(op, grad):
  num_lower = op.inputs[1]
  num_upper = op.inputs[2]
  return (array_ops.matrix_band_part(grad, num_lower, num_upper), None, None)


# Edit Distance has no gradient (but can be used to eval seq2seq or CTC). 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
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 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def _matmul(self, x, transpose_x=False):
    # tf.matmul is defined a * b.
    chol = array_ops.matrix_band_part(self._chol, -1, 0)
    chol_times_x = math_ops.matmul(
        chol, x, transpose_a=True, transpose_b=transpose_x)
    return math_ops.matmul(chol, chol_times_x) 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def _batch_matmul(self, x, transpose_x=False):
    # tf.matmul is defined x * y, so "y" is on the right, not "x".
    chol = array_ops.matrix_band_part(self._chol, -1, 0)
    chol_times_x = math_ops.matmul(
        chol, x, adjoint_a=True, adjoint_b=transpose_x)
    return math_ops.matmul(chol, chol_times_x) 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def _batch_sqrt_matmul(self, x, transpose_x=False):
    chol = array_ops.matrix_band_part(self._chol, -1, 0)
    # tf.batch_matmul is defined x * y, so "y" is on the right, not "x".
    return math_ops.matmul(chol, x, adjoint_b=transpose_x) 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def _sqrt_to_dense(self):
    chol = array_ops.matrix_band_part(self._chol, -1, 0)
    return array_ops.identity(chol) 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def _to_dense(self):
    chol = array_ops.matrix_band_part(self._chol, -1, 0)
    return math_ops.matmul(chol, chol, adjoint_b=True) 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def _preprocess_tril(self, identity_multiplier, diag, tril, event_ndims):
    """Helper to preprocess a lower triangular matrix."""
    tril = array_ops.matrix_band_part(tril, -1, 0)  # Zero out TriU.
    if identity_multiplier is None and diag is None:
      return self._process_matrix(tril, min_rank=2, event_ndims=event_ndims)
    new_diag = array_ops.matrix_diag_part(tril)
    if identity_multiplier is not None:
      new_diag += identity_multiplier
    if diag is not None:
      new_diag += diag
    tril = array_ops.matrix_set_diag(tril, new_diag)
    return self._process_matrix(tril, min_rank=2, event_ndims=event_ndims) 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def _forward(self, x):
    if self._static_event_ndims == 0:
      return math_ops.square(x)
    if self.validate_args:
      is_matrix = check_ops.assert_rank_at_least(x, 2)
      shape = array_ops.shape(x)
      is_square = check_ops.assert_equal(shape[-2], shape[-1])
      x = control_flow_ops.with_dependencies([is_matrix, is_square], x)
    # For safety, explicitly zero-out the upper triangular part.
    x = array_ops.matrix_band_part(x, -1, 0)
    return math_ops.matmul(x, x, adjoint_b=True) 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def random_tril_matrix(shape,
                       dtype,
                       force_well_conditioned=False,
                       remove_upper=True):
  """[batch] lower triangular matrix.

  Args:
    shape:  `TensorShape` or Python `list`.  Shape of the returned matrix.
    dtype:  `TensorFlow` `dtype` or Python dtype
    force_well_conditioned:  Python `bool`. If `True`, returned matrix will have
      eigenvalues with modulus in `(1, 2)`.  Otherwise, eigenvalues are unit
      normal random variables.
    remove_upper:  Python `bool`.
      If `True`, zero out the strictly upper triangle.
      If `False`, the lower triangle of returned matrix will have desired
      properties, but will not not have the strictly upper triangle zero'd out.

  Returns:
    `Tensor` with desired shape and dtype.
  """
  with ops.name_scope("random_tril_matrix"):
    # Totally random matrix.  Has no nice properties.
    tril = random_normal(shape, dtype=dtype)
    if remove_upper:
      tril = array_ops.matrix_band_part(tril, -1, 0)

    # Create a diagonal with entries having modulus in [1, 2].
    if force_well_conditioned:
      maxval = ops.convert_to_tensor(np.sqrt(2.), dtype=dtype.real_dtype)
      diag = random_sign_uniform(
          shape[:-1], dtype=dtype, minval=1., maxval=maxval)
      tril = array_ops.matrix_set_diag(tril, diag)

    return tril 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
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 

# 需要导入模块: from tensorflow.python.ops import array_ops [as 别名]
# 或者: from tensorflow.python.ops.array_ops import matrix_band_part [as 别名]
def _batch_matmul(self, x, transpose_x=False):
    # tf.batch_matmul is defined x * y, so "y" is on the right, not "x".
    chol = array_ops.matrix_band_part(self._chol, -1, 0)
    chol_times_x = math_ops.batch_matmul(
        chol, x, adj_x=True, adj_y=transpose_x)
    return math_ops.batch_matmul(chol, chol_times_x)