def test_resizing_op_adjoint(padding, odl_tspace_impl):
impl = odl_tspace_impl
pad_mode, pad_const = padding
dtypes = [
dt for dt in tensor_space_impl(impl).available_dtypes()
if is_real_floating_dtype(dt)
]
for dtype in dtypes:
space = odl.uniform_discr([0, -1], [1, 1], (4, 5),
dtype=dtype,
impl=impl)
res_space = odl.uniform_discr([0, -1.4], [1.5, 1.4], (6, 7),
dtype=dtype,
impl=impl)
res_op = odl.ResizingOperator(space,
res_space,
pad_mode=pad_mode,
pad_const=pad_const)
if pad_const != 0.0:
with pytest.raises(NotImplementedError):
res_op.adjoint
return
elem = noise_element(space)
res_elem = noise_element(res_space)
inner1 = res_op(elem).inner(res_elem)
inner2 = elem.inner(res_op.adjoint(res_elem))
assert almost_equal(inner1, inner2, places=dtype_places(dtype))
def dspace_type(space, impl, dtype=None):
"""Select the correct corresponding n-tuples space.
Parameters
----------
space : `LinearSpace`
Template space from which to infer an adequate data space. If
it has a `LinearSpace.field` attribute, ``dtype`` must be
consistent with it.
impl : string
Implementation backend for the data space
dtype : `numpy.dtype`, optional
Data type which the space is supposed to use. If ``None`` is
given, the space type is purely determined from ``space`` and
``impl``. Otherwise, it must be compatible with the
field of ``space``.
Returns
-------
stype : type
Space type selected after the space's field, the backend and
the data type
"""
spacetype_map = {
RealNumbers: FN_IMPLS,
ComplexNumbers: FN_IMPLS,
type(None): NTUPLES_IMPLS
}
field_type = type(getattr(space, 'field', None))
if dtype is None:
pass
elif is_real_floating_dtype(dtype):
if field_type is None or field_type == ComplexNumbers:
raise TypeError('real floating data type {!r} requires space '
'field to be of type RealNumbers, got {}'
''.format(dtype, field_type))
elif is_complex_floating_dtype(dtype):
if field_type is None or field_type == RealNumbers:
raise TypeError('complex floating data type {!r} requires space '
'field to be of type ComplexNumbers, got {!r}'
''.format(dtype, field_type))
elif is_scalar_dtype(dtype):
if field_type == ComplexNumbers:
raise TypeError('non-floating data type {!r} requires space field '
'to be of type RealNumbers, got {!r}'.format(
dtype, field_type))
else:
raise TypeError('non-scalar data type {!r} cannot be combined with '
'a `LinearSpace`'.format(dtype))
stype = spacetype_map[field_type].get(impl, None)
if stype is None:
raise NotImplementedError('no corresponding data space available '
'for space {!r} and implementation {!r}'
''.format(space, impl))
return stype
def dspace_type(space, impl, dtype=None):
"""Select the correct corresponding n-tuples space.
Parameters
----------
space : `LinearSpace`
Template space from which to infer an adequate data space. If
it has a `LinearSpace.field` attribute, ``dtype`` must be
consistent with it.
impl : string
Implementation backend for the data space
dtype : `numpy.dtype`, optional
Data type which the space is supposed to use. If ``None`` is
given, the space type is purely determined from ``space`` and
``impl``. Otherwise, it must be compatible with the
field of ``space``.
Returns
-------
stype : type
Space type selected after the space's field, the backend and
the data type
"""
spacetype_map = {RealNumbers: FN_IMPLS,
ComplexNumbers: FN_IMPLS,
type(None): NTUPLES_IMPLS}
field_type = type(getattr(space, 'field', None))
if dtype is None:
pass
elif is_real_floating_dtype(dtype):
if field_type is None or field_type == ComplexNumbers:
raise TypeError('real floating data type {!r} requires space '
'field to be of type RealNumbers, got {}'
''.format(dtype, field_type))
elif is_complex_floating_dtype(dtype):
if field_type is None or field_type == RealNumbers:
raise TypeError('complex floating data type {!r} requires space '
'field to be of type ComplexNumbers, got {!r}'
''.format(dtype, field_type))
elif is_scalar_dtype(dtype):
if field_type == ComplexNumbers:
raise TypeError('non-floating data type {!r} requires space field '
'to be of type RealNumbers, got {!r}'
.format(dtype, field_type))
else:
raise TypeError('non-scalar data type {!r} cannot be combined with '
'a `LinearSpace`'.format(dtype))
stype = spacetype_map[field_type].get(impl, None)
if stype is None:
raise NotImplementedError('no corresponding data space available '
'for space {!r} and implementation {!r}'
''.format(space, impl))
return stype
def tspace_type(space, impl, dtype=None):
"""Select the correct corresponding tensor space.
Parameters
----------
space : `LinearSpace`
Template space from which to infer an adequate tensor space. If
it has a `LinearSpace.field` attribute, ``dtype`` must be
consistent with it.
impl : string
Implementation backend for the tensor space.
dtype : optional
Data type which the space is supposed to use. If ``None`` is
given, the space type is purely determined from ``space`` and
``impl``. Otherwise, it must be compatible with the
field of ``space``.
Returns
-------
stype : type
Space type selected after the space's field, the backend and
the data type.
"""
field_type = type(getattr(space, 'field', None))
if dtype is None:
pass
elif is_real_floating_dtype(dtype):
if field_type is None or field_type == ComplexNumbers:
raise TypeError('real floating data type {!r} requires space '
'field to be of type RealNumbers, got {}'
''.format(dtype, field_type))
elif is_complex_floating_dtype(dtype):
if field_type is None or field_type == RealNumbers:
raise TypeError('complex floating data type {!r} requires space '
'field to be of type ComplexNumbers, got {!r}'
''.format(dtype, field_type))
elif is_numeric_dtype(dtype):
if field_type == ComplexNumbers:
raise TypeError('non-floating data type {!r} requires space field '
'to be of type RealNumbers, got {!r}'.format(
dtype, field_type))
try:
return tensor_space_impl(impl)
except ValueError:
raise NotImplementedError('no corresponding tensor space available '
'for space {!r} and implementation {!r}'
''.format(space, impl))
def test_resizing_op_adjoint(padding, fn_impl):
pad_mode, pad_const = padding
dtypes = [dt for dt in odl.FN_IMPLS[fn_impl].available_dtypes()
if is_real_floating_dtype(dt)]
for dtype in dtypes:
space = odl.uniform_discr([0, -1], [1, 1], (4, 5), dtype=dtype,
impl=fn_impl)
res_space = odl.uniform_discr([0, -1.4], [1.5, 1.4], (6, 7),
dtype=dtype, impl=fn_impl)
res_op = odl.ResizingOperator(space, res_space, pad_mode=pad_mode,
pad_const=pad_const)
if pad_const != 0.0:
with pytest.raises(NotImplementedError):
res_op.adjoint
return
elem = noise_element(space)
res_elem = noise_element(res_space)
inner1 = res_op(elem).inner(res_elem)
inner2 = elem.inner(res_op.adjoint(res_elem))
assert almost_equal(inner1, inner2, places=dtype_places(dtype))
def is_real(self):
"""True if this is a space of real tensors."""
return is_real_floating_dtype(self.dtype)
def dft_postprocess_data(arr, real_grid, recip_grid, shift, axes,
interp, sign='-', op='multiply', out=None):
"""Post-process the Fourier-space data after DFT.
This function multiplies the given data with the separable
function::
q(xi) = exp(+- 1j * dot(x[0], xi)) * s * phi_hat(xi_bar)
where ``x[0]`` and ``s`` are the minimum point and the stride of
the real-space grid, respectively, and ``phi_hat(xi_bar)`` is the FT
of the interpolation kernel. The sign of the exponent depends on the
choice of ``sign``. Note that for ``op='divide'`` the
multiplication with ``s * phi_hat(xi_bar)`` is replaced by a
division with the same array.
In discretized form on the reciprocal grid, the exponential part
of this function becomes an array::
q[k] = exp(+- 1j * dot(x[0], xi[k]))
and the arguments ``xi_bar`` to the interpolation kernel
are the normalized frequencies::
for 'shift=True' : xi_bar[k] = -pi + pi * (2*k) / N
for 'shift=False' : xi_bar[k] = -pi + pi * (2*k+1) / N
See [Pre+2007], Section 13.9 "Computing Fourier Integrals Using
the FFT" for a similar approach.
Parameters
----------
arr : `array-like`
Array to be pre-processed. An array with real data type is
converted to its complex counterpart.
real_grid : uniform `RectGrid`
Real space grid in the transform.
recip_grid : uniform `RectGrid`
Reciprocal grid in the transform
shift : bool or sequence of bools
If ``True``, the grid is shifted by half a stride in the negative
direction in the corresponding axes. The sequence must have the
same length as ``axes``.
axes : int or sequence of ints
Dimensions along which to take the transform. The sequence must
have the same length as ``shifts``.
interp : string or sequence of strings
Interpolation scheme used in the real-space.
sign : {'-', '+'}, optional
Sign of the complex exponent.
op : {'multiply', 'divide'}, optional
Operation to perform with the stride times the interpolation
kernel FT
out : `numpy.ndarray`, optional
Array in which the result is stored. If ``out is arr``, an
in-place modification is performed.
Returns
-------
out : `numpy.ndarray`
Result of the post-processing. If ``out`` was given, the returned
object is a reference to it.
References
----------
[Pre+2007] Press, W H, Teukolsky, S A, Vetterling, W T, and Flannery, B P.
*Numerical Recipes in C - The Art of Scientific Computing* (Volume 3).
Cambridge University Press, 2007.
"""
arr = np.asarray(arr)
if is_real_floating_dtype(arr.dtype):
arr = arr.astype(complex_dtype(arr.dtype))
elif not is_complex_floating_dtype(arr.dtype):
raise ValueError('array data type {} is not a complex floating point '
'data type'.format(dtype_repr(arr.dtype)))
if out is None:
out = arr.copy()
elif out is not arr:
out[:] = arr
if axes is None:
axes = list(range(arr.ndim))
else:
try:
axes = [int(axes)]
except TypeError:
axes = list(axes)
shift_list = normalized_scalar_param_list(shift, length=len(axes),
param_conv=bool)
if sign == '-':
imag = -1j
elif sign == '+':
imag = 1j
else:
raise ValueError("`sign` '{}' not understood".format(sign))
op, op_in = str(op).lower(), op
if op not in ('multiply', 'divide'):
raise ValueError("kernel `op` '{}' not understood".format(op_in))
# Make a list from interp if that's not the case already
try:
# Duck-typed string check
interp + ''
except TypeError:
pass
else:
interp = [str(interp).lower()] * arr.ndim
onedim_arrs = []
for ax, shift, intp in zip(axes, shift_list, interp):
x = real_grid.min_pt[ax]
xi = recip_grid.coord_vectors[ax]
# First part: exponential array
onedim_arr = np.exp(imag * x * xi)
# Second part: interpolation kernel
len_dft = recip_grid.shape[ax]
len_orig = real_grid.shape[ax]
halfcomplex = (len_dft < len_orig)
odd = len_orig % 2
fmin = -0.5 if shift else -0.5 + 1.0 / (2 * len_orig)
if halfcomplex:
# maximum lies around 0, possibly half a cell left or right of it
if shift and odd:
fmax = - 1.0 / (2 * len_orig)
elif not shift and not odd:
fmax = 1.0 / (2 * len_orig)
else:
fmax = 0.0
else: # not halfcomplex
# maximum lies close to 0.5, half or full cell left of it
if shift:
# -0.5 + (N-1)/N = 0.5 - 1/N
fmax = 0.5 - 1.0 / len_orig
else:
# -0.5 + 1/(2*N) + (N-1)/N = 0.5 - 1/(2*N)
fmax = 0.5 - 1.0 / (2 * len_orig)
freqs = np.linspace(fmin, fmax, num=len_dft)
stride = real_grid.stride[ax]
interp_kernel = _interp_kernel_ft(freqs, intp)
interp_kernel *= stride
if op == 'multiply':
onedim_arr *= interp_kernel
else:
onedim_arr /= interp_kernel
onedim_arrs.append(onedim_arr.astype(out.dtype, copy=False))
fast_1d_tensor_mult(out, onedim_arrs, axes=axes, out=out)
return out
def dft_preprocess_data(arr, shift=True, axes=None, sign='-', out=None):
"""Pre-process the real-space data before DFT.
This function multiplies the given data with the separable
function::
p(x) = exp(+- 1j * dot(x - x[0], xi[0]))
where ``x[0]`` and ``xi[0]`` are the minimum coodinates of
the real-space and reciprocal grids, respectively. The sign of
the exponent depends on the choice of ``sign``. In discretized
form, this function becomes an array::
p[k] = exp(+- 1j * k * s * xi[0])
If the reciprocal grid is not shifted, i.e. symmetric around 0,
it is ``xi[0] = pi/s * (-1 + 1/N)``, hence::
p[k] = exp(-+ 1j * pi * k * (1 - 1/N))
For a shifted grid, we have :math:``xi[0] = -pi/s``, thus the
array is given by::
p[k] = (-1)**k
Parameters
----------
arr : `array-like`
Array to be pre-processed. If its data type is a real
non-floating type, it is converted to 'float64'.
shift : bool or or sequence of bools, optional
If ``True``, the grid is shifted by half a stride in the negative
direction. With a sequence, this option is applied separately on
each axis.
axes : int or sequence of ints, optional
Dimensions in which to calculate the reciprocal. The sequence
must have the same length as ``shift`` if the latter is given
as a sequence.
Default: all axes.
sign : {'-', '+'}, optional
Sign of the complex exponent.
out : `numpy.ndarray`, optional
Array in which the result is stored. If ``out is arr``,
an in-place modification is performed. For real data type,
this is only possible for ``shift=True`` since the factors are
complex otherwise.
Returns
-------
out : `numpy.ndarray`
Result of the pre-processing. If ``out`` was given, the returned
object is a reference to it.
Notes
-----
If ``out`` is not specified, the data type of the returned array
is the same as that of ``arr`` except when ``arr`` has real data
type and ``shift`` is not ``True``. In this case, the return type
is the complex counterpart of ``arr.dtype``.
"""
arr = np.asarray(arr)
if not is_scalar_dtype(arr.dtype):
raise ValueError('array has non-scalar data type {}'
''.format(dtype_repr(arr.dtype)))
elif is_real_dtype(arr.dtype) and not is_real_floating_dtype(arr.dtype):
arr = arr.astype('float64')
if axes is None:
axes = list(range(arr.ndim))
else:
try:
axes = [int(axes)]
except TypeError:
axes = list(axes)
shape = arr.shape
shift_list = normalized_scalar_param_list(shift, length=len(axes),
param_conv=bool)
# Make a copy of arr with correct data type if necessary, or copy values.
if out is None:
if is_real_dtype(arr.dtype) and not all(shift_list):
out = np.array(arr, dtype=complex_dtype(arr.dtype), copy=True)
else:
out = arr.copy()
else:
out[:] = arr
if is_real_dtype(out.dtype) and not shift:
raise ValueError('cannot pre-process real input in-place without '
'shift')
if sign == '-':
imag = -1j
elif sign == '+':
imag = 1j
else:
raise ValueError("`sign` '{}' not understood".format(sign))
def _onedim_arr(length, shift):
if shift:
# (-1)^indices
factor = np.ones(length, dtype=out.dtype)
factor[1::2] = -1
else:
factor = np.arange(length, dtype=out.dtype)
factor *= -imag * np.pi * (1 - 1.0 / length)
np.exp(factor, out=factor)
return factor.astype(out.dtype, copy=False)
onedim_arrs = []
for axis, shift in zip(axes, shift_list):
length = shape[axis]
onedim_arrs.append(_onedim_arr(length, shift))
fast_1d_tensor_mult(out, onedim_arrs, axes=axes, out=out)
return out
def dft_postprocess_data(arr, real_grid, recip_grid, shift, axes,
interp, sign='-', op='multiply', out=None):
"""Post-process the Fourier-space data after DFT.
This function multiplies the given data with the separable
function::
q(xi) = exp(+- 1j * dot(x[0], xi)) * s * phi_hat(xi_bar)
where ``x[0]`` and ``s`` are the minimum point and the stride of
the real-space grid, respectively, and ``phi_hat(xi_bar)`` is the FT
of the interpolation kernel. The sign of the exponent depends on the
choice of ``sign``. Note that for ``op='divide'`` the
multiplication with ``s * phi_hat(xi_bar)`` is replaced by a
division with the same array.
In discretized form on the reciprocal grid, the exponential part
of this function becomes an array::
q[k] = exp(+- 1j * dot(x[0], xi[k]))
and the arguments ``xi_bar`` to the interpolation kernel
are the normalized frequencies::
for 'shift=True' : xi_bar[k] = -pi + pi * (2*k) / N
for 'shift=False' : xi_bar[k] = -pi + pi * (2*k+1) / N
See [Pre+2007]_, Section 13.9 "Computing Fourier Integrals Using
the FFT" for a similar approach.
Parameters
----------
arr : `array-like`
Array to be pre-processed. An array with real data type is
converted to its complex counterpart.
real_grid : `RegularGrid`
Real space grid in the transform
recip_grid : `RegularGrid`
Reciprocal grid in the transform
shift : bool or sequence of bools
If ``True``, the grid is shifted by half a stride in the negative
direction in the corresponding axes. The sequence must have the
same length as ``axes``.
axes : int or sequence of ints
Dimensions along which to take the transform. The sequence must
have the same length as ``shifts``.
interp : string or sequence of strings
Interpolation scheme used in the real-space.
sign : {'-', '+'}, optional
Sign of the complex exponent.
op : {'multiply', 'divide'}
Operation to perform with the stride times the interpolation
kernel FT
out : `numpy.ndarray`, optional
Array in which the result is stored. If ``out is arr``, an
in-place modification is performed.
Returns
-------
out : `numpy.ndarray`
Result of the post-processing. If ``out`` was given, the returned
object is a reference to it.
"""
arr = np.asarray(arr)
if is_real_floating_dtype(arr.dtype):
arr = arr.astype(complex_dtype(arr.dtype))
elif not is_complex_floating_dtype(arr.dtype):
raise ValueError('array data type {} is not a complex floating point '
'data type'.format(dtype_repr(arr.dtype)))
if out is None:
out = arr.copy()
elif out is not arr:
out[:] = arr
if axes is None:
axes = list(range(arr.ndim))
else:
try:
axes = [int(axes)]
except TypeError:
axes = list(axes)
shift_list = normalized_scalar_param_list(shift, length=len(axes),
param_conv=bool)
if sign == '-':
imag = -1j
elif sign == '+':
imag = 1j
else:
raise ValueError("`sign` '{}' not understood".format(sign))
op, op_in = str(op).lower(), op
if op not in ('multiply', 'divide'):
raise ValueError("kernel `op` '{}' not understood".format(op_in))
# Make a list from interp if that's not the case already
try:
# Duck-typed string check
interp + ''
except TypeError:
pass
else:
interp = [str(interp).lower()] * arr.ndim
onedim_arrs = []
for ax, shift, intp in zip(axes, shift_list, interp):
x = real_grid.min_pt[ax]
xi = recip_grid.coord_vectors[ax]
# First part: exponential array
onedim_arr = np.exp(imag * x * xi)
# Second part: interpolation kernel
len_dft = recip_grid.shape[ax]
len_orig = real_grid.shape[ax]
halfcomplex = (len_dft < len_orig)
odd = len_orig % 2
fmin = -0.5 if shift else -0.5 + 1.0 / (2 * len_orig)
if halfcomplex:
# maximum lies around 0, possibly half a cell left or right of it
if shift and odd:
fmax = - 1.0 / (2 * len_orig)
elif not shift and not odd:
fmax = 1.0 / (2 * len_orig)
else:
fmax = 0.0
else: # not halfcomplex
# maximum lies close to 0.5, half or full cell left of it
if shift:
# -0.5 + (N-1)/N = 0.5 - 1/N
fmax = 0.5 - 1.0 / len_orig
else:
# -0.5 + 1/(2*N) + (N-1)/N = 0.5 - 1/(2*N)
fmax = 0.5 - 1.0 / (2 * len_orig)
freqs = np.linspace(fmin, fmax, num=len_dft)
stride = real_grid.stride[ax]
if op == 'multiply':
onedim_arr *= stride * _interp_kernel_ft(freqs, intp)
else:
onedim_arr /= stride * _interp_kernel_ft(freqs, intp)
onedim_arrs.append(onedim_arr.astype(out.dtype, copy=False))
fast_1d_tensor_mult(out, onedim_arrs, axes=axes, out=out)
return out
def dft_preprocess_data(arr, shift=True, axes=None, sign='-', out=None):
"""Pre-process the real-space data before DFT.
This function multiplies the given data with the separable
function::
p(x) = exp(+- 1j * dot(x - x[0], xi[0]))
where ``x[0]`` and ``xi[0]`` are the minimum coodinates of
the real-space and reciprocal grids, respectively. The sign of
the exponent depends on the choice of ``sign``. In discretized
form, this function becomes an array::
p[k] = exp(+- 1j * k * s * xi[0])
If the reciprocal grid is not shifted, i.e. symmetric around 0,
it is ``xi[0] = pi/s * (-1 + 1/N)``, hence::
p[k] = exp(-+ 1j * pi * k * (1 - 1/N))
For a shifted grid, we have :math:``xi[0] = -pi/s``, thus the
array is given by::
p[k] = (-1)**k
Parameters
----------
arr : `array-like`
Array to be pre-processed. If its data type is a real
non-floating type, it is converted to 'float64'.
shift : bool or or sequence of bools, optional
If ``True``, the grid is shifted by half a stride in the negative
direction. With a sequence, this option is applied separately on
each axis.
axes : int or sequence of ints, optional
Dimensions in which to calculate the reciprocal. The sequence
must have the same length as ``shift`` if the latter is given
as a sequence.
Default: all axes.
sign : {'-', '+'}, optional
Sign of the complex exponent.
out : `numpy.ndarray`, optional
Array in which the result is stored. If ``out is arr``,
an in-place modification is performed. For real data type,
this is only possible for ``shift=True`` since the factors are
complex otherwise.
Returns
-------
out : `numpy.ndarray`
Result of the pre-processing. If ``out`` was given, the returned
object is a reference to it.
Notes
-----
If ``out`` is not specified, the data type of the returned array
is the same as that of ``arr`` except when ``arr`` has real data
type and ``shift`` is not ``True``. In this case, the return type
is the complex counterpart of ``arr.dtype``.
"""
arr = np.asarray(arr)
if not is_scalar_dtype(arr.dtype):
raise ValueError('array has non-scalar data type {}'
''.format(dtype_repr(arr.dtype)))
elif is_real_dtype(arr.dtype) and not is_real_floating_dtype(arr.dtype):
arr = arr.astype('float64')
if axes is None:
axes = list(range(arr.ndim))
else:
try:
axes = [int(axes)]
except TypeError:
axes = list(axes)
shape = arr.shape
shift_list = normalized_scalar_param_list(shift, length=len(axes),
param_conv=bool)
# Make a copy of arr with correct data type if necessary, or copy values.
if out is None:
if is_real_dtype(arr.dtype) and not all(shift_list):
out = np.array(arr, dtype=complex_dtype(arr.dtype), copy=True)
else:
out = arr.copy()
else:
out[:] = arr
if is_real_dtype(out.dtype) and not shift:
raise ValueError('cannot pre-process real input in-place without '
'shift')
if sign == '-':
imag = -1j
elif sign == '+':
imag = 1j
else:
raise ValueError("`sign` '{}' not understood".format(sign))
def _onedim_arr(length, shift):
if shift:
# (-1)^indices
factor = np.ones(length, dtype=out.dtype)
factor[1::2] = -1
else:
factor = np.arange(length, dtype=out.dtype)
factor *= -imag * np.pi * (1 - 1.0 / length)
np.exp(factor, out=factor)
return factor.astype(out.dtype, copy=False)
onedim_arrs = []
for axis, shift in zip(axes, shift_list):
length = shape[axis]
onedim_arrs.append(_onedim_arr(length, shift))
fast_1d_tensor_mult(out, onedim_arrs, axes=axes, out=out)
return out
