"""Finite-dimensional linear operators."""
from __future__ import annotations
import abc
import functools
from typing import Callable, Optional, Tuple, Union
import numpy as np
import scipy.linalg
import scipy.sparse
from probnum import config # pylint: disable=cyclic-import
from probnum.typing import ArrayLike, DTypeLike, ScalarLike, ShapeLike
import probnum.utils
from . import _vectorize
BinaryOperandType = Union[
"LinearOperator", ScalarLike, np.ndarray, scipy.sparse.spmatrix
]
# pylint: disable="too-many-lines"
class LinearOperator(abc.ABC): # pylint: disable=too-many-instance-attributes
r"""Abstract base class for `matrix-free` finite-dimensional linear operators.
This class provides a way to define finite-dimensional linear operators without
explicitly constructing a matrix representation. Instead it suffices to define a
matrix-matrix product, a :attr:`shape` and a :attr:`dtype`. This avoids unnecessary
memory usage and can often be more convenient to derive.
:class:`LinearOperator`\ s are defined to behave like a :class:`numpy.ndarray` and
thus, they
* have :attr:`shape`, :attr:`dtype`, :attr:`ndim`, and :attr:`size` attributes,
* can be matrix multiplied (:code:`@`) with a :class:`numpy.ndarray` from left and
right, following the same broadcasting rules as :func:`numpy.matmul`,
* can be multiplied (:code:`*`) by a scalar from the left and the right,
* can be added to, subtracted from and matrix multiplied (:code:`@`) with other
:class:`LinearOperator` instances with appropriate :attr:`shape`,
* can be transposed (:attr:`T` or :meth:`transpose`), and they
* can be type-cast (:meth:`astype`).
This is mostly implemented lazily, i.e. the result of these operations is a new,
composite :class:`LinearOperator`, that defers linear operations to the original
operators and combines the results.
Additionally, :class:`LinearOperator`\ s feature
* an efficient :meth:`solve` routine, as well as "lazy" inversion via :meth:`inv`,
* matrix property inference such as :attr:`is_symmetric`,
:attr:`is_positive_definite`, and :attr:`is_lower_triangular`,
* efficient access to matrix factorizations like :meth:`cholesky`, and
* efficient access to derived quantities like the determinant :meth:`det` or
the :meth:`trace`.
Parameters
----------
shape
Matrix dimensions `(M, N)`.
dtype
Data type of the operator.
See Also
--------
aslinop : Transform into a LinearOperator.
Notes
-----
A subclass is only required to implement :meth:`_matmat`. Additionally, other
methods like :meth:`_solve`, :meth:`_inverse`, :meth:`_transpose`,
:meth:`_cholesky`, or :meth:`_det` should be overwritten if more performant
implementations are available.
"""
# pylint: disable=too-many-public-methods
def __init__(
self,
shape: ShapeLike,
dtype: DTypeLike,
):
self.__shape = probnum.utils.as_shape(shape, ndim=2)
# DType
self.__dtype = np.dtype(dtype)
if not np.issubdtype(self.__dtype, np.number):
raise TypeError("The dtype of a linear operator must be numeric.")
if np.issubdtype(self.__dtype, np.complexfloating):
raise TypeError("Linear operators do not support complex dtypes.")
# Matrix properties
self._is_symmetric = None
self._is_lower_triangular = None
self._is_upper_triangular = None
self._is_positive_definite = None
# Caches
self._todense_cache = None
self._rank_cache = None
self._eigvals_cache = None
self._cond_cache = {}
self._det_cache = None
self._logabsdet_cache = None
self._trace_cache = None
self._lu_cache = None
self._cholesky_cache = None
# Property inference
if not self.is_square:
self.is_symmetric = False
@property
def shape(self) -> Tuple[int, int]:
"""Shape of the linear operator.
Defined as a tuple of the output and input dimension of operator.
"""
return self.__shape
@property
def ndim(self) -> int:
"""Number of linear operator dimensions.
Defined analogously to :attr:`numpy.ndarray.ndim`.
"""
return 2
@property
def size(self) -> int:
"""Product of the :attr:`shape` entries.
Defined analogously to :attr:`numpy.ndarray.size`.
"""
return self.__shape[0] * self.__shape[1]
@property
def dtype(self) -> np.dtype:
"""Data type of the linear operator."""
return self.__dtype
@property
def _inexact_dtype(self) -> np.dtype:
if np.issubdtype(self.dtype, np.inexact):
return self.dtype
return np.double
def __repr__(self) -> str:
return (
f"<{self.__class__.__name__} with "
f"shape={self.shape} and "
f"dtype={str(self.dtype)}>"
)
def _apply(self, x: np.ndarray, axis: int) -> np.ndarray:
return np.moveaxis(self @ np.moveaxis(x, axis, -2), -2, axis)
[docs] def __call__(self, x: np.ndarray, axis: Optional[int] = None) -> np.ndarray:
"""Apply the linear operator to an input array along a specified
axis.
Parameters
----------
x : np.ndarray
Input array.
axis : int
Axis along which to apply the linear operator.
Guaranteed to be positive and valid, i.e. ``axis`` is a valid
index into the shape of ``x``, and ``x`` has the correct
shape along ``axis``.
Returns
-------
apply_result : np.ndarray
Array resulting in the application of the linear operator
to ``x`` along ``axis``.
Raises
------
ValueError
If the shape of :code:`x` is invalid.
numpy.AxisError
If the axis argument is not within the valid range.
"""
if axis is not None and (axis < -x.ndim or axis >= x.ndim):
raise np.AxisError(axis, ndim=x.ndim)
if x.ndim == 1:
return self @ x
if x.ndim > 1:
if axis is None:
axis = -1
if axis < 0:
axis += x.ndim
if x.shape[axis] != self.__shape[1]:
raise ValueError(
f"Dimension mismatch. Expected array with {self.__shape[1]} "
f"entries along axis {axis}, but got array with shape {x.shape}."
)
if axis == (x.ndim - 1):
return (self @ x[..., np.newaxis])[..., 0]
if axis == (x.ndim - 2):
return self @ x
return self._apply(x, axis=axis)
raise ValueError("The operand must be at least one dimensional.")
[docs] def solve(self, b: ArrayLike) -> np.ndarray:
"""Solves linear systems :code:`A @ x = b`, where :code:`b` is either a vector
or a (stack of) matrices.
This method broadcasts like :code:`A.inv() @ b`, but it might not produce the
exact same result.
Parameters
----------
b
The right-hand side(s) of the linear system(s). This can either be a vector
or a (stack of) matrices.
Returns
-------
x
The solution(s) of the linear system(s). Depending on the shape of
:code:`b`, :code:`x` is either a vector or a (stack of) matrices.
Raises
------
numpy.linalg.LinAlgError
If the linear operator is non-square or not invertible.
ValueError
If the shape of :code:`b` does not meet the requirements outlined above.
"""
if not self.is_square:
raise np.linalg.LinAlgError("Only square matrices can be inverted.")
b = np.asarray(b)
if b.ndim < 1:
raise ValueError("`b` must be a vector or a (stack of) matrices.")
if b.shape[max(b.ndim - 2, 0)] != self.__shape[0]:
raise ValueError(
f"Dimension mismatch. Expected array with {self.__shape[0]} "
f"entries along axis {max(b.ndim - 2, 1)}, but got array with shape "
f"{b.shape}."
)
if b.ndim == 1:
return self._solve(b[:, None])[:, 0]
return self._solve(b)
@_vectorize.vectorize_matmat(method=True)
def _solve(self, B: np.ndarray) -> np.ndarray:
"""Implementation of the :meth:`solve` method called after input preprocessing.
This method will only be called if the linear operator is square.
The implementation must follow the rules of ``__matmul__`` broadcasting.
When implementing a custom :meth:`_inv` method, then :meth:`_solve` should also
be implemented for performance reasons.
Parameters
----------
B
The right-hand sides of the linear systems. This is guaranteed to be a
(stack of) matrices with appropriate dimensions.
Returns
-------
X
The solutions of the linear system.
Raises
------
numpy.linalg.LinAlgError
The method must throw this exception if the linear operator is not
invertible.
"""
assert B.ndim == 2
if self.is_symmetric:
if self.is_positive_definite is not False:
try:
return scipy.linalg.cho_solve(
(self.cholesky(lower=False).todense(), False),
B,
overwrite_b=False,
)
except np.linalg.LinAlgError:
pass
return scipy.linalg.lu_solve(
self._lu_factor(),
B,
overwrite_b=False,
)
[docs] def astype(
self,
dtype: DTypeLike,
order: str = "K",
casting: str = "unsafe",
subok: bool = True,
copy: bool = True,
) -> "LinearOperator":
"""Cast a linear operator to a different ``dtype``.
Parameters
----------
dtype:
Data type to which the linear operator is cast.
order:
Memory layout order of the result.
casting:
Controls what kind of data casting may occur.
subok:
If True, then sub-classes will be passed-through (default).
False is currently not supported for linear operators.
copy:
Whether to return a new linear operator, even if ``dtype`` is the same.
Raises
------
TypeError
If the linear operator can not be cast to the desired ``dtype`` according to
the given :code:`casting` rule.
NotImplementedError
If :code:`subok` is set to :data:`True`.
"""
dtype = np.dtype(dtype)
if not np.can_cast(self.dtype, dtype, casting=casting):
raise TypeError(
f"Cannot cast linear operator from {self.dtype} to {dtype} "
f"according to the rule {casting}"
)
if not subok:
raise NotImplementedError(
"Setting `subok` to `False` is not supported for linear operators"
)
return self._astype(dtype, order, casting, copy)
def _astype(
self, dtype: np.dtype, order: str, casting: str, copy: bool
) -> "LinearOperator":
if self.dtype == dtype and not copy:
return self
return _TypeCastLinearOperator(self, dtype, order, casting, copy)
def _todense(self) -> np.ndarray:
"""Dense matrix representation of the linear operator.
You may implement this method in a subclass.
Returns
-------
matrix : np.ndarray
Matrix representation of the linear operator.
"""
return self @ np.eye(self.shape[1], dtype=self.__dtype, order="F")
[docs] def todense(self, cache: bool = True) -> np.ndarray:
"""Dense matrix representation of the linear operator.
This method can be computationally very costly depending on the shape of the
linear operator. Use with caution.
Parameters
----------
cache
If this is set to :data:`True`, then the dense matrix representation will
be cached and subsequent calls will return the cached value (even if
:code:`dense` is set to :data:`False` in these subsequent calls).
Returns
-------
matrix : np.ndarray
Matrix representation of the linear operator.
"""
if self._todense_cache is None:
dense = self._todense()
if not cache:
return dense
self._todense_cache = dense
return self._todense_cache
####################################################################################
# Matrix Properties
####################################################################################
@property
def is_square(self) -> bool:
"""Whether input dimension matches output dimension."""
return self.shape[0] == self.shape[1]
@property
def is_symmetric(self) -> Optional[bool]:
"""Whether the ``LinearOperator`` :math:`L` is symmetric, i.e. :math:`L = L^T`.
If this is ``None``, it is unknown whether the operator is symmetric or not.
Only square operators can be symmetric.
Raises
------
ValueError
When setting :attr:`is_symmetric` to :data:`True` on a non-square
:class:`LinearOperator`.
"""
return self._is_symmetric
@is_symmetric.setter
def is_symmetric(self, value: Optional[bool]) -> None:
if value is True and not self.is_square:
raise ValueError("Only square operators can be symmetric.")
self._set_property("symmetric", value)
@property
def is_lower_triangular(self) -> Optional[bool]:
"""Whether the ``LinearOperator`` represents a lower triangular matrix.
If this is ``None``, it is unknown whether the matrix is lower triangular or
not.
"""
return self._is_lower_triangular
@is_lower_triangular.setter
def is_lower_triangular(self, value: Optional[bool]) -> None:
self._set_property("lower_triangular", value)
@property
def is_upper_triangular(self) -> Optional[bool]:
"""Whether the ``LinearOperator`` represents an upper triangular matrix.
If this is ``None``, it is unknown whether the matrix is upper triangular or
not.
"""
return self._is_upper_triangular
@is_upper_triangular.setter
def is_upper_triangular(self, value: Optional[bool]) -> None:
self._set_property("upper_triangular", value)
@property
def is_positive_definite(self) -> Optional[bool]:
"""Whether the ``LinearOperator`` :math:`L \\in \\mathbb{R}^{n \\times n}` is
(strictly) positive-definite, i.e. :math:`x^T L x > 0` for :math:`x \\in \
\\mathbb{R}^n`.
If this is ``None``, it is unknown whether the matrix is positive-definite or
not. Only symmetric operators can be positive-definite.
Raises
------
ValueError
When setting :attr:`is_positive_definite` to :data:`True` while
:attr:`is_symmetric` is :data:`False`.
"""
return self._is_positive_definite
@is_positive_definite.setter
def is_positive_definite(self, value: Optional[bool]) -> None:
if value is True and not self.is_symmetric:
raise ValueError("Only symmetric operators can be positive-definite.")
self._set_property("positive_definite", value)
def _set_property(self, name: str, value: Optional[bool]):
attr_name = f"_is_{name}"
try:
curr_value = getattr(self, attr_name)
except AttributeError as err:
raise AttributeError(
f"The matrix property `{name}` does not exist."
) from err
if curr_value == value:
return
if curr_value is not None:
assert isinstance(curr_value, bool)
raise ValueError(f"Can not change the value of the matrix property {name}.")
if not isinstance(value, bool):
raise TypeError(
f"The value of the matrix property {name} must be a boolean or "
f"`None`, not {type(value)}."
)
setattr(self, attr_name, value)
####################################################################################
# Derived Quantities
####################################################################################
def _rank(self) -> np.intp:
"""Rank of the linear operator.
You may implement this method in a subclass.
"""
return np.linalg.matrix_rank(self.todense(cache=False))
[docs] def rank(self) -> np.intp:
"""Rank of the linear operator."""
if self._rank_cache is None:
self._rank_cache = self._rank()
return self._rank_cache
def _eigvals(self) -> np.ndarray:
"""Eigenvalue spectrum of the linear operator.
You may implement this method in a subclass.
"""
return np.linalg.eigvals(self.todense(cache=False))
[docs] def eigvals(self) -> np.ndarray:
"""Eigenvalue spectrum of the linear operator.
Raises
------
numpy.linalg.LinAlgError
If :meth:`eigvals` is called on a non-square operator.
"""
if self._eigvals_cache is None:
if not self.is_square:
raise np.linalg.LinAlgError(
"Eigenvalues are only defined for square operators"
)
self._eigvals_cache = self._eigvals()
self._eigvals_cache.setflags(write=False)
return self._eigvals_cache
def _cond(
self, p: Optional[Union[None, int, str, np.floating]] = None
) -> np.floating:
"""Compute the condition number of the linear operator.
The condition number of the linear operator with respect to the ``p`` norm. It
measures how much the solution :math:`x` of the linear system :math:`Ax=b`
changes with respect to small changes in :math:`b`.
The linear operator is guaranteed to be square.
You may implement this method in a subclass.
Parameters
----------
p : {None, 1, , 2, , inf, 'fro'}, optional
Order of the norm:
======= ============================
p norm for matrices
======= ============================
None 2-norm, computed directly via singular value decomposition
'fro' Frobenius norm
np.inf max(sum(abs(x), axis=1))
1 max(sum(abs(x), axis=0))
2 2-norm (largest sing. value)
======= ============================
Returns
-------
cond :
The condition number of the linear operator. May be infinite.
"""
return np.linalg.cond(self.todense(cache=False), p=p)
[docs] def cond(
self, p: Optional[Union[None, int, str, np.floating]] = None
) -> np.floating:
"""Compute the condition number of the linear operator.
The condition number of the linear operator with respect to the ``p`` norm. It
measures how much the solution :math:`x` of the linear system :math:`Ax=b`
changes with respect to small changes in :math:`b`.
Parameters
----------
p : {None, 1, , 2, , inf, 'fro'}, optional
Order of the norm:
======= ============================
p norm for matrices
======= ============================
None 2-norm, computed directly via singular value decomposition
'fro' Frobenius norm
np.inf max(sum(abs(x), axis=1))
1 max(sum(abs(x), axis=0))
2 2-norm (largest sing. value)
======= ============================
Returns
-------
cond :
The condition number of the linear operator. May be infinite.
Raises
------
numpy.linalg.LinAlgError
If :meth:`cond` is called on a non-square matrix.
"""
if p not in self._cond_cache:
if not self.is_square:
raise np.linalg.LinAlgError(
"The condition number is only defined for square operators"
)
self._cond_cache[p] = self._cond(p)
return self._cond_cache[p]
@functools.cached_property
def _slogdet(self) -> Tuple[np.inexact, np.floating]:
return np.linalg.slogdet(self.todense(cache=False))
def _det(self) -> np.inexact:
"""Determinant of the linear operator.
The linear operator is guaranteed to be square.
You may implement this method in a subclass.
Returns
-------
det :
The determinant of the linear operator.
"""
sign, logabsdet = self._slogdet
return sign * np.exp(logabsdet)
[docs] def det(self) -> np.inexact:
"""Determinant of the linear operator.
Returns
-------
det :
The determinant of the linear operator.
Raises
------
numpy.linalg.LinAlgError
If :meth:`det` is called on a non-square matrix.
"""
if self._det_cache is None:
if not self.is_square:
raise np.linalg.LinAlgError(
"The determinant is only defined for square operators"
)
self._det_cache = self._det()
return self._det_cache
def _logabsdet(self) -> np.floating:
"""Log absolute determinant of the linear operator.
The linear operator is guaranteed to be square.
You may implement this method in a subclass.
Returns
-------
logabsdet :
The log absolute determinant of the linear operator.
"""
_, logabsdet = self._slogdet
return logabsdet
[docs] def logabsdet(self) -> np.floating:
"""Log absolute determinant of the linear operator.
Returns
-------
logabsdet :
The log absolute determinant of the linear operator.
Raises
------
numpy.linalg.LinAlgError
If :meth:`logabsdet` is called on a non-square matrix.
"""
if self._logabsdet_cache is None:
if not self.is_square:
raise np.linalg.LinAlgError(
"The determinant is only defined for square operators"
)
self._logabsdet_cache = self._logabsdet()
return self._logabsdet_cache
def _trace(self) -> np.number:
r"""Trace of the linear operator.
Computes the trace of a square linear operator :math:`\text{tr}(A) =
\sum_{i-1}^n A_{ii}`.
The linear operator is guaranteed to be square.
You may implement this method in a subclass.
Returns
-------
trace : float
Trace of the linear operator.
"""
vec = np.zeros(self.shape[1], dtype=self.dtype)
vec[0] = 1
trace = (self @ vec)[0]
vec[0] = 0
for i in range(1, self.shape[0]):
vec[i] = 1
trace += (self @ vec)[i]
vec[i] = 0
return trace
[docs] def trace(self) -> np.number:
r"""Trace of the linear operator.
Computes the trace of a square linear operator :math:`\text{tr}(A) =
\sum_{i-1}^n A_{ii}`.
Returns
-------
trace : float
Trace of the linear operator.
Raises
------
numpy.linalg.LinAlgError
If :meth:`trace` is called on a non-square matrix.
"""
if self._trace_cache is None:
if not self.is_square:
raise np.linalg.LinAlgError(
"The trace is only defined for square operators."
)
self._trace_cache = self._trace()
return self._trace_cache
####################################################################################
# Matrix Decompositions
####################################################################################
[docs] def cholesky(self, lower: bool = True) -> LinearOperator:
r"""Computes a Cholesky decomposition of the :class:`LinearOperator`.
The Cholesky decomposition of a symmetric positive-definite matrix :math:`A \in
\mathbb{R}^{n \times n}` is given by :math:`A = L L^T`, where the unique
Cholesky factor :math:`L \in \mathbb{R}^{n \times n}` of :math:`A` is a lower
triangular matrix with a positive diagonal.
As a side-effect, this method will set the value of :attr:`is_positive_definite`
to :obj:`True`, if the computation of the Cholesky factorization succeeds.
Otherwise, :attr:`is_positive_definite` will be set to :obj:`False`.
The result of this computation will be cached. If :meth:`cholesky` is first
called with ``lower=True`` first and afterwards with ``lower=False`` or
vice-versa, the method simply returns the transpose of the cached value.
Parameters
----------
lower :
If this is set to :obj:`False`, this method computes and returns the
upper triangular Cholesky factor :math:`U = L^T`, for which :math:`A = U^T
U`. By default (:obj:`True`), the method computes the lower triangular
Cholesky factor :math:`L`.
Returns
-------
cholesky_factor :
The lower or upper Cholesky factor of the :class:`LinearOperator`, depending
on the value of the parameter ``lower``. The result will have its properties
:attr:`is_upper_triangular`\ /:attr:`is_lower_triangular` set accordingly.
Raises
------
numpy.linalg.LinAlgError
If the :class:`LinearOperator` is not symmetric, i.e. if
:attr:`is_symmetric` is not set to :obj:`True`.
numpy.linalg.LinAlgError
If the :class:`LinearOperator` is not positive definite.
"""
if not self.is_symmetric:
raise np.linalg.LinAlgError(
"The Cholesky decomposition is only defined for symmetric matrices."
)
if self.is_positive_definite is False:
raise np.linalg.LinAlgError("The linear operator is not positive definite.")
if self._cholesky_cache is None:
try:
self._cholesky_cache = self._cholesky(lower)
self.is_positive_definite = True
except np.linalg.LinAlgError as err:
self.is_positive_definite = False
raise err
if lower:
self._cholesky_cache.is_lower_triangular = True
else:
self._cholesky_cache.is_upper_triangular = True
upper = not lower
if (lower and self._cholesky_cache.is_lower_triangular) or (
upper and self._cholesky_cache.is_upper_triangular
):
return self._cholesky_cache
assert (
self._cholesky_cache.is_lower_triangular
or self._cholesky_cache.is_upper_triangular
)
return self._cholesky_cache.T
def _cholesky(self, lower: bool) -> LinearOperator:
return Matrix(
scipy.linalg.cholesky(
self.todense(), lower=lower, overwrite_a=False, check_finite=True
)
)
def _lu_factor(self):
"""This is a modified version of the original implementation in SciPy:
https://github.com/scipy/scipy/blob/v1.7.1/scipy/linalg/decomp_lu.py#L15-L84
because the SciPy implementation does not raise an exception if the matrix is
singular.
"""
if self._lu_cache is None:
from scipy.linalg.lapack import ( # pylint: disable=no-name-in-module,import-outside-toplevel
get_lapack_funcs,
)
a = np.asarray_chkfinite(self.todense())
(getrf,) = get_lapack_funcs(("getrf",), (a,))
lu, piv, info = getrf(a, overwrite_a=False)
if info < 0:
raise ValueError(
f"illegal value in argument {-info} of internal getrf (lu_factor)"
)
if info > 0:
raise np.linalg.LinAlgError(
f"Diagonal number {info} is exactly zero. Singular matrix."
)
self._lu_cache = lu, piv
return self._lu_cache
####################################################################################
# Unary Arithmetic
####################################################################################
def __neg__(self) -> "LinearOperator":
# pylint: disable=import-outside-toplevel,cyclic-import
from ._arithmetic import NegatedLinearOperator
return NegatedLinearOperator(self)
def _transpose(self) -> "LinearOperator":
"""Transpose of this linear operator.
You may implement this method in a subclass.
Returns
-------
transpose : LinearOperator
Transpose of this linear operator, which is again
a LinearOperator.
"""
# This does not need caching, since the `TransposeLinearOperator`
# only accesses quantities (particularly `todense`), which are
# cached inside the original `LinearOperator`.
return TransposedLinearOperator(self)
@property
def T(self) -> "LinearOperator":
"""Transpose of the linear operator."""
if self.is_symmetric:
return self
transposed = self._transpose()
transposed.is_upper_triangular = self.is_lower_triangular
transposed.is_lower_triangular = self.is_upper_triangular
transposed.is_symmetric = self.is_symmetric
transposed.is_positive_definite = self.is_positive_definite
return transposed
[docs] def transpose(self, *axes: Union[int, Tuple[int]]) -> "LinearOperator":
"""Transpose this linear operator.
Can be abbreviated self.T instead of self.transpose().
Parameters
----------
*axes
Permutation of the axes of the :class:`LinearOperator`.
Raises
------
ValueError
If the given axis indices do not constitute a valid permutation of the axes.
numpy.AxisError
If the axis indices are out of bounds.
"""
if len(axes) > 0:
if len(axes) == 1 and isinstance(axes[0], tuple):
axes = axes[0]
if len(axes) != 2:
raise ValueError(
f"The given axes {axes} don't match the linear operator with shape "
f"{self.shape}."
)
axes_int = []
for axis in axes:
axis = int(axis)
if not -2 <= axis <= 1:
raise np.AxisError(axis, ndim=2)
if axis < 0:
axis += 2
axes_int.append(axis)
axes = tuple(axes_int)
if axes == (0, 1):
return self
if axes == (1, 0):
return self.T
raise ValueError("Cannot transpose a linear operator along repeated axes.")
return self.T
def _inverse(self) -> "LinearOperator":
"""Inverse of this linear operator.
The linear operator is guaranteed to be square.
You may implement this method in a subclass.
Returns
-------
inv : LinearOperator
Inverse of this linear operator, which is again
a LinearOperator.
"""
# This does not need caching, since the `_InverseLinearOperator` only accesses
# quantities (particularly matrix decompositions), which are cached inside the
# original `LinearOperator`.
return _InverseLinearOperator(self)
[docs] def inv(self) -> "LinearOperator":
"""Inverse of the linear operator.
Returns
-------
inv : LinearOperator
Inverse of this linear operator, which is again
a LinearOperator.
Raises
------
numpy.linalg.LinAlgError
If :meth:`inv` is called on a non-square linear operator.
"""
if not self.is_square:
raise np.linalg.LinAlgError(
"Inverses are only defined on square linear operators."
)
return self._inverse()
[docs] def symmetrize(self) -> LinearOperator:
"""Compute or approximate the closest symmetric :class:`LinearOperator` in the
Frobenius norm.
The closest symmetric matrix to a given square matrix :math:`A` in the Frobenius
norm is given by
.. math::
\\operatorname{sym}(A) := \\frac{1}{2} (A + A^T).
However, for efficiency reasons, it is preferrable to approximate this operator
in some cases. For example, a Kronecker product :math:`K = A \\otimes B` is more
efficiently symmetrized by means of
.. math::
:nowrap:
\\begin{equation}
\\operatorname{sym}(A) \\otimes \\operatorname{sym}(B)
= \\operatorname{sym}(K) + \\frac{1}{2} \\left(
\\frac{1}{2} \\left(
A \\otimes B^T + A^T \\otimes B
\\right) - \\operatorname{sym}(K)
\\right).
\\end{equation}
Returns
-------
symmetrized_linop :
The closest symmetric :class:`LinearOperator` in the Frobenius norm, or an
approximation, which makes a reasonable trade-off between accuracy and
efficiency (see above). The resulting :class:`LinearOperator` will have its
:attr:`is_symmetric` property set to :obj:`True`.
Raises
------
numpy.linalg.LinAlgError
If this method is called on a non-square :class:`LinearOperator`.
"""
if not self.is_square:
raise np.linalg.LinAlgError("A non-square operator can not be symmetrized.")
if self.is_symmetric:
return self
linop_sym = self._symmetrize()
linop_sym.is_symmetric = True
return linop_sym
def _symmetrize(self) -> LinearOperator:
return 0.5 * (self + self.T)
####################################################################################
# Binary Arithmetic
####################################################################################
__array_ufunc__ = None
"""
This prevents numpy from calling elementwise arithmetic operations allowing
expressions like `y = np.array([1, 1]) + linop` to call the arithmetic operations
defined by `LinearOperator` instead of elementwise. Thus no array of
`LinearOperator`s but a `LinearOperator` with the correct shape is returned.
"""
def __add__(self, other: BinaryOperandType) -> "LinearOperator":
from ._arithmetic import add # pylint: disable=import-outside-toplevel
return add(self, other)
def __radd__(self, other: BinaryOperandType) -> "LinearOperator":
from ._arithmetic import add # pylint: disable=import-outside-toplevel
return add(other, self)
def __sub__(self, other: BinaryOperandType) -> "LinearOperator":
from ._arithmetic import sub # pylint: disable=import-outside-toplevel
return sub(self, other)
def __rsub__(self, other: BinaryOperandType) -> "LinearOperator":
from ._arithmetic import sub # pylint: disable=import-outside-toplevel
return sub(other, self)
def __mul__(self, other: BinaryOperandType) -> "LinearOperator":
from ._arithmetic import mul # pylint: disable=import-outside-toplevel
return mul(self, other)
def __rmul__(self, other: BinaryOperandType) -> "LinearOperator":
from ._arithmetic import mul # pylint: disable=import-outside-toplevel
return mul(other, self)
def _is_type_shape_dtype_equal(self, other: "LinearOperator") -> bool:
return (
isinstance(self, type(other))
and self.shape == other.shape
and self.dtype == other.dtype
)
@abc.abstractmethod
def _matmul(self, x: np.ndarray) -> np.ndarray:
"""Matrix multiplication.
Performs the operation `M = self @ x` where `self` is
an MxN linear operator and `x` is a stack of matrices
of shape `(..., N, K)`.
The shapes are guaranteed to be correct.
You must implement this method in a subclass.
Parameters
----------
x :
A stack of matrices of shape `(..., N, K)`.
Returns
-------
M :
A `np.ndarray` of shape `(..., M, K)` that is the result of
`M = self @ x`.
"""
def __matmul__(
self, other: BinaryOperandType
) -> Union["LinearOperator", np.ndarray]:
"""Matrix-vector multiplication.
Performs the operation `y = self @ x` where `self` is
an MxN linear operator and `x` is a 1-d array or random variable.
Parameters
----------
x :
An array or `RandomVariable` with shape `(N,)` or `(N, 1)`.
Returns
-------
y :
A `np.matrix` or `np.ndarray` or `RandomVariable` with
shape `(M,)` or `(M, 1)`,depending on the type and
shape of the x argument.
Raises
------
ValueError
If the shape of :code:`other` is invalid.
Notes
-----
This matvec wraps the user-specified matvec routine or overridden
_matvec method to ensure that y has the correct shape and type.
"""
if isinstance(other, np.ndarray):
x = other
M, N = self.shape
if x.ndim == 1 and x.shape == (N,):
y = self._matmul(x[:, np.newaxis])[:, 0]
assert y.ndim == 1
assert y.shape == (M,)
elif x.ndim > 1 and x.shape[-2] == N:
y = self._matmul(x)
assert y.ndim > 1
assert y.shape == x.shape[:-2] + (M, x.shape[-1])
else:
raise ValueError(
f"Dimension mismatch. Expected operand of shape ({N},) or "
f"(..., {N}, K), but got {x.shape}."
)
return y
# pylint: disable=import-outside-toplevel,cyclic-import
from ._arithmetic import matmul
return matmul(self, other)
def __rmatmul__(
self, other: BinaryOperandType
) -> Union["LinearOperator", np.ndarray]:
if isinstance(other, np.ndarray):
x = other
M, N = self.shape
if x.ndim >= 1 and x.shape[-1] == M:
y = (self.T)(x, axis=-1) # pylint: disable=not-callable
else:
raise ValueError(
f"Dimension mismatch. Expected operand of shape (..., {M}), but "
f"got {x.shape}."
)
assert y.ndim >= 1
assert y.shape[-1] == N
assert y.shape[:-1] == x.shape[:-1]
return y
# pylint: disable=import-outside-toplevel,cyclic-import
from ._arithmetic import matmul
return matmul(other, self)
####################################################################################
# Automatic `mat{vec,mat}`` to `matmul` Vectorization
####################################################################################
broadcast_matvec = staticmethod(_vectorize.vectorize_matvec)
broadcast_matmat = staticmethod(_vectorize.vectorize_matmat)
class LambdaLinearOperator( # pylint: disable=too-many-instance-attributes
LinearOperator
):
r"""Convenience subclass of LinearOperator that lets you pass
implementations of its methods as parameters instead of
overriding them in a subclass.
``shape``, ``dtype`` and ``matmul`` must be passed, the other
parameters are optional.
Parameters
----------
shape
Matrix dimensions `(M, N)`.
dtype
Data type of the operator.
matmul
Callable which computes the matrix-matrix product :math:`y = A V`, where
:math:`A` is the linear operator and :math:`V` is an :math:`N \times K` matrix.
The callable must support broadcasted matrix products, i.e. the argument
:math:`V` might also be a stack of matrices in which case the broadcasting rules
of :func:`np.matmul` must apply.
Note that the argument to this callable is guaranteed to have at least two
dimensions.
apply
Callable which implements the application of the linear operator to an input
array along a specified axis.
todense
Callable which returns a dense matrix representation of the linear operator as a
:class:`np.ndarray`. The output of this function must be equivalent to the
output of :code:`A.matmat(np.eye(N, dtype=A.dtype))`.
transpose
Callable which returns a LinearOperator that corresponds to the
transpose of this linear operator.
inverse
Callable which returns a LinearOperator that corresponds to the
inverse of this linear operator.
rank
Callable which returns the rank of this linear operator.
eigvals
Callable which returns the eigenvalues of this linear operator.
cond
Callable which returns the condition number of this
linear operator.
det
Callable which returns the determinant of this linear operator.
logabsdet
Callable which returns the log absolute determinant of this
linear operator.
trace
Callable which returns the trace of this linear operator.
Examples
--------
>>> import numpy as np
>>> from probnum.linops import LambdaLinearOperator, LinearOperator
>>> @LinearOperator.broadcast_matvec
... def mv(v):
... return np.array([2 * v[0] - v[1], 3 * v[1]])
>>> A = LambdaLinearOperator(shape=(2, 2), dtype=np.float_, matmul=mv)
>>> A
<LambdaLinearOperator with shape=(2, 2) and dtype=float64>
>>> A @ np.array([1., 2.])
array([0., 6.])
>>> A @ np.ones(2)
array([1., 3.])
"""
def __init__(
self,
shape: ShapeLike,
dtype: DTypeLike,
*,
matmul: Callable[[np.ndarray], np.ndarray],
apply: Callable[[np.ndarray, int], np.ndarray] = None,
solve: Callable[[np.ndarray], np.ndarray] = None,
todense: Optional[Callable[[], np.ndarray]] = None,
transpose: Optional[Callable[[], LinearOperator]] = None,
inverse: Optional[Callable[[], LinearOperator]] = None,
rank: Optional[Callable[[], np.intp]] = None,
eigvals: Optional[Callable[[], np.ndarray]] = None,
cond: Optional[
Callable[[Optional[Union[None, int, str, np.floating]]], np.floating]
] = None,
det: Optional[Callable[[], np.inexact]] = None,
logabsdet: Optional[Callable[[], np.floating]] = None,
trace: Optional[Callable[[], np.number]] = None,
):
super().__init__(shape, dtype)
self._matmul_fn = matmul # (self @ x)
self._apply_fn = apply # __call__
self._solve_fn = solve
self._todense_fn = todense
self._transpose_fn = transpose
self._inverse_fn = inverse
# Derived quantities
self._rank_fn = rank
self._eigvals_fn = eigvals
self._cond_fn = cond
self._det_fn = det
self._logabsdet_fn = logabsdet
self._trace_fn = trace
def _matmul(self, x: np.ndarray) -> np.ndarray:
return self._matmul_fn(x)
def _apply(self, x: np.ndarray, axis: int) -> np.ndarray:
if self._apply_fn is None:
return super()._apply(x, axis=axis)
return self._apply_fn(x, axis)
def _solve(self, B: np.ndarray) -> np.ndarray:
if self._solve_fn is None:
return super()._solve(B)
return self._solve_fn(B)
def _todense(self) -> np.ndarray:
if self._todense_fn is None:
return super()._todense()
return self._todense_fn()
def _transpose(self) -> LinearOperator:
if self._transpose_fn is None:
return super()._transpose()
return self._transpose_fn()
def _inverse(self) -> LinearOperator:
if self._inverse_fn is None:
return super()._inverse()
return self._inverse_fn()
def _rank(self) -> np.intp:
if self._rank_fn is None:
return super()._rank()
return self._rank_fn()
def _eigvals(self) -> np.ndarray:
if self._eigvals_fn is None:
return super()._eigvals()
return self._eigvals_fn()
def _cond(
self, p: Optional[Union[None, int, str, np.floating]] = None
) -> np.floating:
if self._cond_fn is None:
return super()._cond(p)
return self._cond_fn(p)
def _det(self) -> np.inexact:
if self._det_fn is None:
return super()._det()
return self._det_fn()
def _logabsdet(self) -> np.floating:
if self._logabsdet_fn is None:
return super()._logabsdet()
return self._logabsdet_fn()
def _trace(self) -> np.number:
if self._trace_fn is None:
return super()._trace()
return self._trace_fn()
class TransposedLinearOperator(LambdaLinearOperator):
"""Transposition of a linear operator."""
def __init__(
self,
linop: LinearOperator,
matmul: Optional[Callable[[np.ndarray], np.ndarray]] = None,
):
self._linop = linop
if matmul is None:
# Setting `cache=True` here does not allocate any extra memory, since the
# transpose of `self._linop.todense(cache=True)` is just a view of the
# original array
matmul = lambda x: self.todense(cache=True) @ x
super().__init__(
shape=(self._linop.shape[1], self._linop.shape[0]),
dtype=self._linop.dtype,
matmul=matmul,
todense=lambda: self._linop.todense(cache=True).T,
transpose=lambda: self._linop,
inverse=lambda: self._linop.inv().T,
rank=self._linop.rank,
det=self._linop.det,
logabsdet=self._linop.logabsdet,
trace=self._linop.trace,
)
def _astype(
self, dtype: np.dtype, order: str, casting: str, copy: bool
) -> LinearOperator:
return self._linop.astype(dtype, order=order, casting=casting, copy=copy).T
def __repr__(self) -> str:
return f"Transpose of {self._linop}"
def _cholesky(self, lower: bool = True) -> LinearOperator:
return super().cholesky(lower)
class _InverseLinearOperator(LambdaLinearOperator):
def __init__(self, linop: LinearOperator):
if not linop.is_square:
raise np.linalg.LinAlgError("Only square operators can be inverted.")
self._linop = linop
super().__init__(
shape=self._linop.shape,
dtype=self._linop._inexact_dtype,
matmul=self._linop.solve,
transpose=lambda: TransposedLinearOperator(
self,
matmul=self._tmatmul,
),
inverse=lambda: self._linop,
det=lambda: 1 / self._linop.det(),
logabsdet=lambda: -self._linop.logabsdet(),
)
# Matrix properties
self.is_symmetric = self._linop.is_symmetric
self.is_positive_definite = self._linop.is_positive_definite
def __repr__(self) -> str:
return f"Inverse of {self._linop}"
@LinearOperator.broadcast_matmat(method=True)
def _tmatmul(self, B: ArrayLike) -> np.ndarray:
assert B.ndim == 2
if self.is_symmetric:
if self.is_positive_definite is not False:
try:
return scipy.linalg.cho_solve(
(self._linop.cholesky(lower=False).todense(), False),
B,
overwrite_b=False,
)
except np.linalg.LinAlgError:
pass
return scipy.linalg.lu_solve(
self._linop._lu_factor(), # pylint: disable=protected-access
B,
trans=1,
overwrite_b=False,
)
class _TypeCastLinearOperator(LambdaLinearOperator):
def __init__(
self,
linop: LinearOperator,
dtype: DTypeLike,
order: str = "K",
casting: str = "unsafe",
copy: bool = True,
):
self._linop = linop
dtype = np.dtype(dtype)
if not np.can_cast(self._linop.dtype, dtype, casting=casting):
raise TypeError(
f"Cannot cast linear operator from {self._linop.dtype} to {dtype} "
f"according to the rule {casting}"
)
super().__init__(
self._linop.shape,
dtype,
matmul=lambda x: (self._linop @ x).astype(
np.result_type(self.dtype, x.dtype), copy=False
),
apply=lambda x, axis: self._linop(x, axis=axis).astype(
np.result_type(self.dtype, x.dtype), copy=False
),
solve=lambda b: self.inv() @ b,
todense=lambda: self._linop.todense(cache=False).astype(
dtype, order=order, copy=copy
),
transpose=lambda: self._linop.T.astype(dtype),
inverse=lambda: self._linop.inv().astype(self._inexact_dtype),
rank=self._linop.rank,
eigvals=lambda: self._linop.eigvals().astype(self._inexact_dtype),
cond=lambda p: self._linop.cond(p=p).astype(self._inexact_dtype),
det=lambda: self._linop.det().astype(self._inexact_dtype),
logabsdet=lambda: self._linop.logabsdet().astype(self._inexact_dtype),
trace=lambda: self._linop.trace().astype(dtype),
)
def _astype(
self, dtype: np.dtype, order: str, casting: str, copy: bool
) -> LinearOperator:
if self.dtype == dtype and not copy:
return self
if dtype == self._linop.dtype and not copy:
return self._linop
return _TypeCastLinearOperator(self, dtype, order, casting, copy)
class Matrix(LambdaLinearOperator):
"""A linear operator defined via a matrix.
Parameters
----------
A :
The explicit matrix.
"""
def __init__(self, A: Union[ArrayLike, scipy.sparse.spmatrix]):
if isinstance(A, scipy.sparse.spmatrix):
self.A = A
matmul = LinearOperator.broadcast_matmat(lambda x: self.A @ x)
todense = self.A.toarray
trace = lambda: self.A.diagonal().sum()
else:
self.A = np.asarray(A)
self.A.setflags(write=False)
matmul = lambda x: self.A @ x
todense = lambda: self.A
trace = lambda: np.trace(self.A)
super().__init__(
self.A.shape,
self.A.dtype,
matmul=matmul,
todense=todense,
trace=trace,
)
def _transpose(self) -> "Matrix":
return Matrix(self.A.T)
def _astype(
self, dtype: np.dtype, order: str, casting: str, copy: bool
) -> "LinearOperator":
if self.dtype == dtype and not copy:
return self
if isinstance(self.A, np.ndarray):
A_astype = self.A.astype(dtype, order=order, casting=casting, copy=copy)
else:
assert isinstance(self.A, scipy.sparse.spmatrix)
A_astype = self.A.astype(dtype, casting=casting, copy=copy)
return Matrix(A_astype)
def _matmul_matrix(self, other: "Matrix") -> "Matrix":
if not config.lazy_matrix_matrix_matmul:
if not self.shape[1] == other.shape[0]:
raise ValueError(f"Matmul shape mismatch {self.shape} x {other.shape}")
return Matrix(A=self.A @ other.A)
return NotImplemented
def __eq__(self, other: LinearOperator) -> bool:
if not self._is_type_shape_dtype_equal(other):
return False
return np.all(self.A == other.A)
def __neg__(self) -> "Matrix":
return Matrix(-self.A)
def _symmetrize(self) -> LinearOperator:
return Matrix(0.5 * (self.A + self.A.T))
class Identity(LambdaLinearOperator):
"""The identity operator.
Parameters
----------
shape :
Shape of the identity operator.
"""
def __init__(
self,
shape: ShapeLike,
dtype: DTypeLike = np.double,
):
shape = probnum.utils.as_shape(shape)
if len(shape) == 1:
shape = 2 * shape
elif len(shape) != 2:
raise ValueError("The shape of a linear operator must be two-dimensional.")
if shape[0] != shape[1]:
raise np.linalg.LinAlgError("An identity operator must be square.")
super().__init__(
shape,
dtype,
matmul=lambda x: x.astype(np.result_type(self.dtype, x.dtype), copy=False),
apply=lambda x, axis: x.astype(
np.result_type(self.dtype, x.dtype), copy=False
),
todense=lambda: np.identity(self.shape[0], dtype=dtype),
transpose=lambda: self,
inverse=lambda: self,
rank=lambda: np.intp(shape[0]),
eigvals=lambda: np.ones(shape[0], dtype=self._inexact_dtype),
cond=self._cond,
det=lambda: probnum.utils.as_numpy_scalar(1.0, dtype=self._inexact_dtype),
logabsdet=lambda: probnum.utils.as_numpy_scalar(
0.0, dtype=self._inexact_dtype
),
trace=lambda: probnum.utils.as_numpy_scalar(
self.shape[0], dtype=self.dtype
),
)
# Matrix properties
self.is_symmetric = True
self.is_lower_triangular = True
self.is_upper_triangular = True
self.is_positive_definite = True
def _solve(self, B: np.ndarray) -> np.ndarray:
return B
def _cond(
self, p: Optional[Union[None, int, str, np.floating]] = None
) -> np.floating:
if p is None or p in (2, 1, np.inf, -2, -1, -np.inf):
return probnum.utils.as_numpy_scalar(1.0, dtype=self._inexact_dtype)
if p == "fro":
return probnum.utils.as_numpy_scalar(
self.shape[0], dtype=self._inexact_dtype
)
return super()._cond(p)
def _astype(
self, dtype: np.dtype, order: str, casting: str, copy: bool
) -> "Identity":
if dtype == self.dtype and not copy:
return self
return Identity(self.shape, dtype=dtype)
def __eq__(self, other: LinearOperator) -> bool:
return self._is_type_shape_dtype_equal(other)
def _cholesky(self, lower: bool = True) -> LinearOperator:
return self
class Selection(LambdaLinearOperator):
"""Indexing into a vector at one or multiple indices, represented as a
:class:`LinearOperator`.
Parameters
----------
indices:
Indices to select.
shape:
Shape of the linear operator.
dtype:
Data type of the linear operator.
"""
def __init__(self, indices, shape, dtype=np.double):
if np.ndim(indices) > 1:
raise ValueError(
"Selection LinOp expects an integer or (1D) iterable of "
f"integers. Received {type(indices)} with shape {np.shape(indices)}."
)
if shape[0] > shape[1]:
raise ValueError(
f"Invalid shape {shape} for Selection LinOp. If the "
"output-dimension (shape[0]) is larger than the input-dimension "
"(shape[1]), consider using `Embedding`."
)
self._indices = probnum.utils.as_shape(indices)
assert len(self._indices) == shape[0]
super().__init__(
dtype=dtype,
shape=shape,
matmul=lambda x: _selection_matmul(self.indices, x),
todense=self._todense,
transpose=lambda: Embedding(
take_indices=np.arange(len(self._indices)),
put_indices=self._indices,
shape=(self.shape[1], self.shape[0]),
),
)
@property
def indices(self) -> Tuple[int]:
"""Indices which will be selected when applying the linear operator to a
vector."""
return self._indices
def _todense(self):
res = np.eye(self.shape[1], self.shape[1])
return _selection_matmul(self.indices, res)
def _selection_matmul(indices, M):
return np.take(M, indices=indices, axis=-2)
class Embedding(LambdaLinearOperator):
"""Embeds a vector into a higher-dimensional space by writing its entries to the
given indices of the result vector."""
def __init__(
self, take_indices, put_indices, shape, fill_value=0.0, dtype=np.double
):
if np.ndim(take_indices) > 1:
raise ValueError(
"Embedding LinOp expects an integer or (1D) iterable of "
f"integers. Received {type(take_indices)} with shape "
f"{np.shape(take_indices)}."
)
if np.ndim(put_indices) > 1:
raise ValueError(
"Embedding LinOp expects an integer or (1D) iterable of "
f"integers. Received {type(put_indices)} with shape "
f"{np.shape(put_indices)}."
)
if shape[0] < shape[1]:
raise ValueError(
f"Invalid shape {shape} for Embedding LinOp. If the "
"output-dimension (shape[0]) is smaller than the input-dimension "
"(shape[1]), consider using `Selection`."
)
self._take_indices = probnum.utils.as_shape(take_indices)
self._put_indices = probnum.utils.as_shape(put_indices)
self._fill_value = fill_value
super().__init__(
dtype=dtype,
shape=shape,
matmul=lambda x: _embedding_matmul(self, x),
todense=self._todense,
transpose=lambda: Selection(
indices=put_indices, shape=(self.shape[1], self.shape[0])
),
)
def _todense(self):
return self.T.todense().T
def _embedding_matmul(embedding, M):
# pylint: disable=protected-access
res_shape = np.array(M.shape)
res_shape[-2] = embedding.shape[0]
res = np.full(shape=tuple(res_shape), fill_value=embedding._fill_value)
take_from_M = M[..., np.array(embedding._take_indices), :]
res[..., np.array(embedding._put_indices), :] = take_from_M
return res