def linsolve(self, A, b):
"""
Solve linear equation set ``Ax = b`` and returns the solutions in a 1-D array.
This function performs both symbolic and numeric factorizations every time, and can be slower than
``Solver.solve``.
Parameters
----------
A
Sparse matrix
b
RHS of the equation
Returns
-------
The solution in a 1-D np array.
"""
if self.sparselib == 'umfpack':
try:
umfpack.linsolve(A, b)
except ArithmeticError:
logger.error('Singular matrix. Case is not solvable')
return np.ravel(b)
elif self.sparselib == 'klu':
try:
klu.linsolve(A, b)
except ArithmeticError:
logger.error('Singular matrix. Case is not solvable')
return np.ravel(b)
elif self.sparselib in ('spsolve', 'cupy'):
ccs = A.CCS
size = A.size
data = np.array(ccs[2]).reshape((-1,))
indices = np.array(ccs[1]).reshape((-1,))
indptr = np.array(ccs[0]).reshape((-1,))
A = csc_matrix((data, indices, indptr), shape=size)
if self.sparselib == 'spsolve':
x = spsolve(A, b)
return np.ravel(x)
elif self.sparselib == 'cupy':
# delayed import for startup speed
import cupy as cp # NOQA
from cupyx.scipy.sparse import csc_matrix as csc_cu # NOQA
from cupyx.scipy.sparse.linalg.solve import lsqr as cu_lsqr # NOQA
cu_A = csc_cu(A)
cu_b = cp.array(np.array(b).reshape((-1,)))
x = cu_lsqr(cu_A, cu_b)
return np.ravel(cp.asnumpy(x[0]))
def linsolve(self, A, b):
try:
klu.linsolve(A, b)
except ArithmeticError:
logger.error('Singular matrix. Case is not solvable')
return np.ravel(b)