def __init__(self, diag: Iterable, lmat: Iterable):
"""
Parameters
----------
diag : Iterable
Diagonal vector.
lmat : Iterable
Low rank matrix.
Raises
------
ValueError
If length of ``diag`` not match with number of rows of ``lmat``.
ValueError
If there are non-positive numbers in ``diag``.
"""
diag = utils.to_numpy(diag, ndim=(1, ))
lmat = utils.to_numpy(lmat, ndim=(2, ))
if diag.size != lmat.shape[0]:
raise ValueError("`diag` and `lmat` size not match.")
if any(diag <= 0.0):
raise ValueError("`diag` must be all positive.")
self.diag = diag
self.lmat = lmat
self.dsize = self.diag.size
self.lrank = min(self.lmat.shape)
self.sdiag = np.sqrt(self.diag)
self.ilmat = ILMat(self.lmat / self.sdiag[:, np.newaxis])
def invdot(self, x: Iterable) -> ndarray:
"""
Inverse dot product with vector or matrix
Parameters
----------
x : Iterable
Vector or matrix
Returns
-------
ndarray
"""
x = utils.to_numpy(x, ndim=(1, 2))
return x + (self._u * self._w) @ (self._u.T @ x)
def __init__(self, lmat: Iterable):
"""
Parameters
----------
lmat : Iterable
Raises
------
ValueError
When ``lmat`` is not a matrix.
"""
self.lmat = utils.to_numpy(lmat, ndim=(2, ))
self.dsize = self.lmat.shape[0]
self.lrank = min(self.lmat.shape)
self._u, s, _ = np.linalg.svd(self.lmat, full_matrices=False)
self._v = s**2
self._w = -self._v / (1 + self._v)
def invdot(self, x: Iterable) -> ndarray:
"""
Inverse dot product with vector or matrix
Parameters
----------
x : Iterable
Vector or matrix
Returns
-------
ndarray
"""
x = utils.to_numpy(x, ndim=(1, 2))
x = (x.T / self.sdiag).T
x = self.ilmat.invdot(x)
x = (x.T / self.sdiag).T
return x
def test_to_numpy(array):
nparray = utils.to_numpy(array)
assert isinstance(nparray, np.ndarray)
def test_to_numpy_ndim(array):
with pytest.raises(ValueError):
utils.to_numpy(array, ndim=(1, ))