273 lines
8.0 KiB
Python
273 lines
8.0 KiB
Python
"""Utilities to deal with sympy.Matrix, numpy and scipy.sparse."""
|
|
|
|
from sympy.core.expr import Expr
|
|
from sympy.core.numbers import I
|
|
from sympy.core.singleton import S
|
|
from sympy.matrices.matrices import MatrixBase
|
|
from sympy.matrices import eye, zeros
|
|
from sympy.external import import_module
|
|
|
|
__all__ = [
|
|
'numpy_ndarray',
|
|
'scipy_sparse_matrix',
|
|
'sympy_to_numpy',
|
|
'sympy_to_scipy_sparse',
|
|
'numpy_to_sympy',
|
|
'scipy_sparse_to_sympy',
|
|
'flatten_scalar',
|
|
'matrix_dagger',
|
|
'to_sympy',
|
|
'to_numpy',
|
|
'to_scipy_sparse',
|
|
'matrix_tensor_product',
|
|
'matrix_zeros'
|
|
]
|
|
|
|
# Conditionally define the base classes for numpy and scipy.sparse arrays
|
|
# for use in isinstance tests.
|
|
|
|
np = import_module('numpy')
|
|
if not np:
|
|
class numpy_ndarray:
|
|
pass
|
|
else:
|
|
numpy_ndarray = np.ndarray # type: ignore
|
|
|
|
scipy = import_module('scipy', import_kwargs={'fromlist': ['sparse']})
|
|
if not scipy:
|
|
class scipy_sparse_matrix:
|
|
pass
|
|
sparse = None
|
|
else:
|
|
sparse = scipy.sparse
|
|
scipy_sparse_matrix = sparse.spmatrix # type: ignore
|
|
|
|
|
|
def sympy_to_numpy(m, **options):
|
|
"""Convert a SymPy Matrix/complex number to a numpy matrix or scalar."""
|
|
if not np:
|
|
raise ImportError
|
|
dtype = options.get('dtype', 'complex')
|
|
if isinstance(m, MatrixBase):
|
|
return np.array(m.tolist(), dtype=dtype)
|
|
elif isinstance(m, Expr):
|
|
if m.is_Number or m.is_NumberSymbol or m == I:
|
|
return complex(m)
|
|
raise TypeError('Expected MatrixBase or complex scalar, got: %r' % m)
|
|
|
|
|
|
def sympy_to_scipy_sparse(m, **options):
|
|
"""Convert a SymPy Matrix/complex number to a numpy matrix or scalar."""
|
|
if not np or not sparse:
|
|
raise ImportError
|
|
dtype = options.get('dtype', 'complex')
|
|
if isinstance(m, MatrixBase):
|
|
return sparse.csr_matrix(np.array(m.tolist(), dtype=dtype))
|
|
elif isinstance(m, Expr):
|
|
if m.is_Number or m.is_NumberSymbol or m == I:
|
|
return complex(m)
|
|
raise TypeError('Expected MatrixBase or complex scalar, got: %r' % m)
|
|
|
|
|
|
def scipy_sparse_to_sympy(m, **options):
|
|
"""Convert a scipy.sparse matrix to a SymPy matrix."""
|
|
return MatrixBase(m.todense())
|
|
|
|
|
|
def numpy_to_sympy(m, **options):
|
|
"""Convert a numpy matrix to a SymPy matrix."""
|
|
return MatrixBase(m)
|
|
|
|
|
|
def to_sympy(m, **options):
|
|
"""Convert a numpy/scipy.sparse matrix to a SymPy matrix."""
|
|
if isinstance(m, MatrixBase):
|
|
return m
|
|
elif isinstance(m, numpy_ndarray):
|
|
return numpy_to_sympy(m)
|
|
elif isinstance(m, scipy_sparse_matrix):
|
|
return scipy_sparse_to_sympy(m)
|
|
elif isinstance(m, Expr):
|
|
return m
|
|
raise TypeError('Expected sympy/numpy/scipy.sparse matrix, got: %r' % m)
|
|
|
|
|
|
def to_numpy(m, **options):
|
|
"""Convert a sympy/scipy.sparse matrix to a numpy matrix."""
|
|
dtype = options.get('dtype', 'complex')
|
|
if isinstance(m, (MatrixBase, Expr)):
|
|
return sympy_to_numpy(m, dtype=dtype)
|
|
elif isinstance(m, numpy_ndarray):
|
|
return m
|
|
elif isinstance(m, scipy_sparse_matrix):
|
|
return m.todense()
|
|
raise TypeError('Expected sympy/numpy/scipy.sparse matrix, got: %r' % m)
|
|
|
|
|
|
def to_scipy_sparse(m, **options):
|
|
"""Convert a sympy/numpy matrix to a scipy.sparse matrix."""
|
|
dtype = options.get('dtype', 'complex')
|
|
if isinstance(m, (MatrixBase, Expr)):
|
|
return sympy_to_scipy_sparse(m, dtype=dtype)
|
|
elif isinstance(m, numpy_ndarray):
|
|
if not sparse:
|
|
raise ImportError
|
|
return sparse.csr_matrix(m)
|
|
elif isinstance(m, scipy_sparse_matrix):
|
|
return m
|
|
raise TypeError('Expected sympy/numpy/scipy.sparse matrix, got: %r' % m)
|
|
|
|
|
|
def flatten_scalar(e):
|
|
"""Flatten a 1x1 matrix to a scalar, return larger matrices unchanged."""
|
|
if isinstance(e, MatrixBase):
|
|
if e.shape == (1, 1):
|
|
e = e[0]
|
|
if isinstance(e, (numpy_ndarray, scipy_sparse_matrix)):
|
|
if e.shape == (1, 1):
|
|
e = complex(e[0, 0])
|
|
return e
|
|
|
|
|
|
def matrix_dagger(e):
|
|
"""Return the dagger of a sympy/numpy/scipy.sparse matrix."""
|
|
if isinstance(e, MatrixBase):
|
|
return e.H
|
|
elif isinstance(e, (numpy_ndarray, scipy_sparse_matrix)):
|
|
return e.conjugate().transpose()
|
|
raise TypeError('Expected sympy/numpy/scipy.sparse matrix, got: %r' % e)
|
|
|
|
|
|
# TODO: Move this into sympy.matricies.
|
|
def _sympy_tensor_product(*matrices):
|
|
"""Compute the kronecker product of a sequence of SymPy Matrices.
|
|
"""
|
|
from sympy.matrices.expressions.kronecker import matrix_kronecker_product
|
|
|
|
return matrix_kronecker_product(*matrices)
|
|
|
|
|
|
def _numpy_tensor_product(*product):
|
|
"""numpy version of tensor product of multiple arguments."""
|
|
if not np:
|
|
raise ImportError
|
|
answer = product[0]
|
|
for item in product[1:]:
|
|
answer = np.kron(answer, item)
|
|
return answer
|
|
|
|
|
|
def _scipy_sparse_tensor_product(*product):
|
|
"""scipy.sparse version of tensor product of multiple arguments."""
|
|
if not sparse:
|
|
raise ImportError
|
|
answer = product[0]
|
|
for item in product[1:]:
|
|
answer = sparse.kron(answer, item)
|
|
# The final matrices will just be multiplied, so csr is a good final
|
|
# sparse format.
|
|
return sparse.csr_matrix(answer)
|
|
|
|
|
|
def matrix_tensor_product(*product):
|
|
"""Compute the matrix tensor product of sympy/numpy/scipy.sparse matrices."""
|
|
if isinstance(product[0], MatrixBase):
|
|
return _sympy_tensor_product(*product)
|
|
elif isinstance(product[0], numpy_ndarray):
|
|
return _numpy_tensor_product(*product)
|
|
elif isinstance(product[0], scipy_sparse_matrix):
|
|
return _scipy_sparse_tensor_product(*product)
|
|
|
|
|
|
def _numpy_eye(n):
|
|
"""numpy version of complex eye."""
|
|
if not np:
|
|
raise ImportError
|
|
return np.array(np.eye(n, dtype='complex'))
|
|
|
|
|
|
def _scipy_sparse_eye(n):
|
|
"""scipy.sparse version of complex eye."""
|
|
if not sparse:
|
|
raise ImportError
|
|
return sparse.eye(n, n, dtype='complex')
|
|
|
|
|
|
def matrix_eye(n, **options):
|
|
"""Get the version of eye and tensor_product for a given format."""
|
|
format = options.get('format', 'sympy')
|
|
if format == 'sympy':
|
|
return eye(n)
|
|
elif format == 'numpy':
|
|
return _numpy_eye(n)
|
|
elif format == 'scipy.sparse':
|
|
return _scipy_sparse_eye(n)
|
|
raise NotImplementedError('Invalid format: %r' % format)
|
|
|
|
|
|
def _numpy_zeros(m, n, **options):
|
|
"""numpy version of zeros."""
|
|
dtype = options.get('dtype', 'float64')
|
|
if not np:
|
|
raise ImportError
|
|
return np.zeros((m, n), dtype=dtype)
|
|
|
|
|
|
def _scipy_sparse_zeros(m, n, **options):
|
|
"""scipy.sparse version of zeros."""
|
|
spmatrix = options.get('spmatrix', 'csr')
|
|
dtype = options.get('dtype', 'float64')
|
|
if not sparse:
|
|
raise ImportError
|
|
if spmatrix == 'lil':
|
|
return sparse.lil_matrix((m, n), dtype=dtype)
|
|
elif spmatrix == 'csr':
|
|
return sparse.csr_matrix((m, n), dtype=dtype)
|
|
|
|
|
|
def matrix_zeros(m, n, **options):
|
|
""""Get a zeros matrix for a given format."""
|
|
format = options.get('format', 'sympy')
|
|
if format == 'sympy':
|
|
return zeros(m, n)
|
|
elif format == 'numpy':
|
|
return _numpy_zeros(m, n, **options)
|
|
elif format == 'scipy.sparse':
|
|
return _scipy_sparse_zeros(m, n, **options)
|
|
raise NotImplementedError('Invaild format: %r' % format)
|
|
|
|
|
|
def _numpy_matrix_to_zero(e):
|
|
"""Convert a numpy zero matrix to the zero scalar."""
|
|
if not np:
|
|
raise ImportError
|
|
test = np.zeros_like(e)
|
|
if np.allclose(e, test):
|
|
return 0.0
|
|
else:
|
|
return e
|
|
|
|
|
|
def _scipy_sparse_matrix_to_zero(e):
|
|
"""Convert a scipy.sparse zero matrix to the zero scalar."""
|
|
if not np:
|
|
raise ImportError
|
|
edense = e.todense()
|
|
test = np.zeros_like(edense)
|
|
if np.allclose(edense, test):
|
|
return 0.0
|
|
else:
|
|
return e
|
|
|
|
|
|
def matrix_to_zero(e):
|
|
"""Convert a zero matrix to the scalar zero."""
|
|
if isinstance(e, MatrixBase):
|
|
if zeros(*e.shape) == e:
|
|
e = S.Zero
|
|
elif isinstance(e, numpy_ndarray):
|
|
e = _numpy_matrix_to_zero(e)
|
|
elif isinstance(e, scipy_sparse_matrix):
|
|
e = _scipy_sparse_matrix_to_zero(e)
|
|
return e
|