def __init__(self, uspace, dspace, restr=None, ext=None, **kwargs):
"""Abstract initialization method.
Intended to be called by subclasses for proper type checking
and setting of attributes.
Parameters
----------
uspace : :class:`~odl.LinearSpace`
The (abstract) space to be discretized
dspace : :class:`~odl.space.base_ntuples.FnBase`
Data space providing containers for the values of a
discretized object. Its
:attr:`~odl.space.base_ntuples.FnBase.field` attribute
must be the same as ``uspace.field``.
restr : :class:`~odl.Operator`, linear, optional
Operator mapping a :attr:`RawDiscretization.uspace` element
to a :attr:`RawDiscretization.dspace` element. Must satisfy
``restr.domain == uspace``, ``restr.range == dspace``
ext : :class:`~odl.Operator`, linear, optional
Operator mapping a :attr:`RawDiscretization.dspace` element
to a :attr:`RawDiscretization.uspace` element. Must satisfy
``ext.domain == dspace``, ``ext.range == uspace``.
"""
super().__init__(uspace, dspace, restr, ext, **kwargs)
FnBase.__init__(self, dspace.size, dspace.dtype)
if not isinstance(uspace, LinearSpace):
raise TypeError('undiscretized space {} not a `LinearSpace` '
'instance.'.format(uspace))
if not isinstance(dspace, FnBase):
raise TypeError('data space {} not an `FnBase` instance.'
''.format(dspace))
if uspace.field != dspace.field:
raise ValueError('fields {} and {} of the undiscretized and '
'data spaces, resp., are not equal.'
''.format(uspace.field, dspace.field))
if restr is not None:
if not isinstance(restr, Operator):
raise TypeError('restriction operator {} is not a '
'`Operator` instance.'.format(restr))
if not restr.is_linear:
raise TypeError('restriction operator {} is not '
'linear'.format(restr))
if ext is not None:
if not isinstance(ext, Operator):
raise TypeError('extension operator {} is not a '
'`Operator` instance.'.format(ext))
if not ext.is_linear:
raise TypeError('extension operator {} is not '
'linear'.format(ext))
def __init__(self, uspace, dspace, sampling=None, interpol=None):
"""Abstract initialization method.
Intended to be called by subclasses for proper type checking
and setting of attributes.
Parameters
----------
uspace : `LinearSpace`
The (abstract) space to be discretized
dspace : `FnBase`
Data space providing containers for the values of a
discretized object. Its `FnBase.field` attribute
must be the same as ``uspace.field``.
sampling : `Operator`, linear, optional
Operator mapping a `RawDiscretization.uspace` element
to a `RawDiscretization.dspace` element. Must satisfy
``sampling.domain == uspace``, ``sampling.range == dspace``
interpol : `Operator`, linear, optional
Operator mapping a `RawDiscretization.dspace` element
to a `RawDiscretization.uspace` element. Must satisfy
``interpol.domain == dspace``, ``interpol.range == uspace``.
"""
RawDiscretization.__init__(self, uspace, dspace, sampling, interpol)
FnBase.__init__(self, dspace.size, dspace.dtype)
if not isinstance(uspace, LinearSpace):
raise TypeError('undiscretized space {!r} not a LinearSpace '
'instance.'.format(uspace))
if not isinstance(dspace, FnBase):
raise TypeError('data space {!r} not an FnBase instance.'
''.format(dspace))
if uspace.field != dspace.field:
raise ValueError('fields {} and {} of the undiscretized and '
'data spaces, resp., are not equal.'
''.format(uspace.field, dspace.field))
if sampling is not None:
if not isinstance(sampling, Operator):
raise TypeError('sampling operator {!r} is not an '
'Operator instance.'.format(sampling))
if not sampling.is_linear:
raise TypeError('sampling operator {!r} is not '
'linear'.format(sampling))
if interpol is not None:
if not isinstance(interpol, Operator):
raise TypeError('interpolation operator {!r} is not an '
'Operator instance.'.format(interpol))
if not interpol.is_linear:
raise TypeError('interpolation operator {!r} is not '
'linear'.format(interpol))
def __init__(self, uspace, dspace, sampling=None, interpol=None):
"""Abstract initialization method.
Intended to be called by subclasses for proper type checking
and setting of attributes.
Parameters
----------
uspace : `LinearSpace`
The (abstract) space to be discretized
dspace : `FnBase`
Data space providing containers for the values of a
discretized object. Its `FnBase.field` attribute
must be the same as ``uspace.field``.
sampling : `Operator`, linear, optional
Operator mapping a `DiscretizedSet.uspace` element
to a `DiscretizedSet.dspace` element. Must satisfy
``sampling.domain == uspace``, ``sampling.range == dspace``
interpol : `Operator`, linear, optional
Operator mapping a `DiscretizedSet.dspace` element
to a `DiscretizedSet.uspace` element. Must satisfy
``interpol.domain == dspace``, ``interpol.range == uspace``.
"""
DiscretizedSet.__init__(self, uspace, dspace, sampling, interpol)
FnBase.__init__(self, dspace.size, dspace.dtype)
if not isinstance(uspace, LinearSpace):
raise TypeError('`uspace` {!r} not a LinearSpace '
'instance'.format(uspace))
if not isinstance(dspace, FnBase):
raise TypeError('`dspace` {!r} not an FnBase instance'
''.format(dspace))
if uspace.field != dspace.field:
raise ValueError('fields {} and {} of the undiscretized and '
'data spaces, resp., are not equal'
''.format(uspace.field, dspace.field))
if sampling is not None and not sampling.is_linear:
raise TypeError('`sampling` {!r} is not linear'
''.format(sampling))
if interpol is not None and not interpol.is_linear:
raise TypeError('`interpol` {!r} is not linear'
''.format(interpol))
def __init__(self, size):
FnBase.__init__(self, size, np.float)
def __init__(self, size):
FnBase.__init__(self, size, np.float)
def __init__(self, size, dtype='float32', **kwargs):
"""Initialize a new instance.
Parameters
----------
size : positive `int`
The number of dimensions of the space
dtype : `object`
The data type of the storage array. Can be provided in any
way the `numpy.dtype` function understands, most notably
as built-in type, as one of NumPy's internal datatype
objects or as string.
Only scalar data types are allowed.
weight : optional
Use weighted inner product, norm, and dist. The following
types are supported as ``weight``:
`FnWeightingBase` :
Use this weighting as-is. Compatibility with this
space's elements is not checked during init.
`float` :
Weighting by a constant
`array-like` :
Weighting by a vector (1-dim. array, corresponds to
a diagonal matrix). Note that the array is stored in
main memory, which results in slower space functions
due to a copy during evaluation.
`CudaFnVector` :
same as 1-dim. array-like, except that copying is
avoided if the ``dtype`` of the vector is the
same as this space's ``dtype``.
Default: no weighting
This option cannot be combined with ``dist``, ``norm``
or ``inner``.
exponent : positive `float`, optional
Exponent of the norm. For values other than 2.0, no
inner product is defined.
This option is ignored if ``dist``, ``norm`` or
``inner`` is given.
Default: 2.0
dist : `callable`, optional
The distance function defining a metric on the space.
It must accept two `FnVector` arguments and
fulfill the following mathematical conditions for any
three vectors ``x, y, z``:
- ``dist(x, y) >= 0``
- ``dist(x, y) = 0`` if and only if ``x = y``
- ``dist(x, y) = dist(y, x)``
- ``dist(x, y) <= dist(x, z) + dist(z, y)``
This option cannot be combined with ``weight``,
``norm`` or ``inner``.
norm : `callable`, optional
The norm implementation. It must accept an
`FnVector` argument, return a `float` and satisfy the
following conditions for all vectors ``x, y`` and scalars
``s``:
- ``||x|| >= 0``
- ``||x|| = 0`` if and only if ``x = 0``
- ``||s * x|| = |s| * ||x||``
- ``||x + y|| <= ||x|| + ||y||``
By default, ``norm(x)`` is calculated as ``inner(x, x)``.
This option cannot be combined with ``weight``,
``dist`` or ``inner``.
inner : `callable`, optional
The inner product implementation. It must accept two
`FnVector` arguments, return a element from
the field of the space (real or complex number) and
satisfy the following conditions for all vectors
``x, y, z`` and scalars ``s``:
- ``<x, y> = conj(<y, x>)``
- ``<s*x + y, z> = s * <x, z> + <y, z>``
- ``<x, x> = 0`` if and only if ``x = 0``
This option cannot be combined with ``weight``,
``dist`` or ``norm``.
"""
FnBase.__init__(self, size, dtype)
CudaNtuples.__init__(self, size, dtype)
dist = kwargs.pop('dist', None)
norm = kwargs.pop('norm', None)
inner = kwargs.pop('inner', None)
weight = kwargs.pop('weight', None)
exponent = kwargs.pop('exponent', 2.0)
# Check validity of option combination (3 or 4 out of 4 must be None)
if sum(x is None for x in (dist, norm, inner, weight)) < 3:
raise ValueError('invalid combination of options `weight`, '
'`dist`, `norm` and `inner`')
if weight is not None:
if isinstance(weight, WeightingBase):
self._weighting = weight
elif np.isscalar(weight):
self._weighting = CudaFnConstWeighting(
weight, exponent=exponent)
elif isinstance(weight, CudaFnVector):
self._weighting = CudaFnVectorWeighting(
weight, exponent=exponent)
else:
# Must make a CudaFnVector from the array
weight = self.element(np.asarray(weight))
if weight.ndim == 1:
self._weighting = CudaFnVectorWeighting(
weight, exponent=exponent)
else:
raise ValueError('invalid weight argument {!r}'
''.format(weight))
elif dist is not None:
self._weighting = CudaFnCustomDist(dist)
elif norm is not None:
self._weighting = CudaFnCustomNorm(norm)
elif inner is not None:
# Use fast dist implementation
self._weighting = CudaFnCustomInnerProduct(
inner, dist_using_inner=True)
else: # all None -> no weighing
self._weighting = CudaFnNoWeighting(exponent)
def __init__(self, size, dtype='float32', **kwargs):
"""Initialize a new instance.
Parameters
----------
size : positive `int`
The number of dimensions of the space
dtype : `object`
The data type of the storage array. Can be provided in any
way the `numpy.dtype` function understands, most notably
as built-in type, as one of NumPy's internal datatype
objects or as string.
Only scalar data types are allowed.
weighting : optional
Use weighted inner product, norm, and dist. The following
types are supported as ``weight``:
`FnWeightingBase` :
Use this weighting as-is. Compatibility with this
space's elements is not checked during init.
`float` :
Weighting by a constant
`array-like` :
Weighting by a vector (1-dim. array, corresponds to
a diagonal matrix). Note that the array is stored in
main memory, which results in slower space functions
due to a copy during evaluation.
`CudaFnVector` :
same as 1-dim. array-like, except that copying is
avoided if the ``dtype`` of the vector is the
same as this space's ``dtype``.
Default: no weighting
This option cannot be combined with ``dist``, ``norm``
or ``inner``.
exponent : positive `float`, optional
Exponent of the norm. For values other than 2.0, no
inner product is defined.
This option is ignored if ``dist``, ``norm`` or
``inner`` is given.
Default: 2.0
dist : `callable`, optional
The distance function defining a metric on the space.
It must accept two `FnVector` arguments and
fulfill the following mathematical conditions for any
three vectors ``x, y, z``:
- ``dist(x, y) >= 0``
- ``dist(x, y) = 0`` if and only if ``x = y``
- ``dist(x, y) = dist(y, x)``
- ``dist(x, y) <= dist(x, z) + dist(z, y)``
This option cannot be combined with ``weight``,
``norm`` or ``inner``.
norm : `callable`, optional
The norm implementation. It must accept an
`FnVector` argument, return a `float` and satisfy the
following conditions for all vectors ``x, y`` and scalars
``s``:
- ``||x|| >= 0``
- ``||x|| = 0`` if and only if ``x = 0``
- ``||s * x|| = |s| * ||x||``
- ``||x + y|| <= ||x|| + ||y||``
By default, ``norm(x)`` is calculated as ``inner(x, x)``.
This option cannot be combined with ``weight``,
``dist`` or ``inner``.
inner : `callable`, optional
The inner product implementation. It must accept two
`FnVector` arguments, return a element from
the field of the space (real or complex number) and
satisfy the following conditions for all vectors
``x, y, z`` and scalars ``s``:
- ``<x, y> = conj(<y, x>)``
- ``<s*x + y, z> = s * <x, z> + <y, z>``
- ``<x, x> = 0`` if and only if ``x = 0``
This option cannot be combined with ``weight``,
``dist`` or ``norm``.
"""
FnBase.__init__(self, size, dtype)
CudaNtuples.__init__(self, size, dtype)
dist = kwargs.pop('dist', None)
norm = kwargs.pop('norm', None)
inner = kwargs.pop('inner', None)
weighting = kwargs.pop('weighting', None)
exponent = kwargs.pop('exponent', 2.0)
# Check validity of option combination (3 or 4 out of 4 must be None)
if sum(x is None for x in (dist, norm, inner, weighting)) < 3:
raise ValueError('invalid combination of options `weight`, '
'`dist`, `norm` and `inner`')
if weighting is not None:
if isinstance(weighting, Weighting):
self.__weighting = weighting
elif np.isscalar(weighting):
self.__weighting = CudaFnConstWeighting(weighting,
exponent=exponent)
elif isinstance(weighting, CudaFnVector):
self.__weighting = CudaFnArrayWeighting(weighting,
exponent=exponent)
else:
# Must make a CudaFnVector from the array
weighting = self.element(np.asarray(weighting))
if weighting.ndim == 1:
self.__weighting = CudaFnArrayWeighting(weighting,
exponent=exponent)
else:
raise ValueError('invalid weighting argument {!r}'
''.format(weighting))
elif dist is not None:
self.__weighting = CudaFnCustomDist(dist)
elif norm is not None:
self.__weighting = CudaFnCustomNorm(norm)
elif inner is not None:
# Use fast dist implementation
self.__weighting = CudaFnCustomInner(inner, dist_using_inner=True)
else: # all None -> no weighing
self.__weighting = CudaFnNoWeighting(exponent)
