WIP: improve repr by a lot

- add helper `_separators` to determine string
  separators that keep a `repr` string within a
  given line width
- split repr string rendering into two functions,
  one to determine the parts to be printed and
  one to compile the final string (with separators)
- `attribute_repr_string` helper to class attributes
- `element_repr_string` helper to print space
  elements
- make repr printing of product space elements more
  consistent
- improve default repr of `Operator` (now actually
  good by default, but unspecific)
- use Numpy's print options as a single point to
  configure printing
- add doctests to lots of `__repr__` methods

Further miscellaneous changes:
- fix bug of `uniform_discr_fromdiscr` ignoring dtype
- sort imports using the `isort` tool (mode `-m 4`)
- add `asweighted` to base_tensors to make weighted
  and unweighted spaces compatible in tensor operators
- remove base classes for operator and adjoint in
  favor of inner adjoint classes (less code)
- make most default operators take `domain` and `range`
  and implement the corresponding adjoints/inverses etc.
- implement `noise_array` for product spaces that
  are not power spaces
- remove the `almost_equal` test util
- improve a number of doctests (use `odl.` calls,
  better names etc.)
- make comparison of arrays in tests faster
- implement broadcasting of shapes (M, N) and (N,)
  in product spaces
- remove some remaining doc glitches, mostly `FnBase`

TODO:
- go through rest of the library
- replace `almost_equal` by `pytest.approx`
- tests for pspace broadcasting
- make `uniform_discr_fromdiscr` respect weighting
- fix failing unit tests

Author: Holger Kohr

.gitignore

@@ -90,4 +90,7 @@ target/
 *.ipynb
 .ipynb_checkpoints
 
+# VS Code files
+.vscode/
+
 # Documentation build files

odl/README.md

@@ -21,7 +21,7 @@ This is a brief description of the content of each submodule, see the individual
 * [phantom](phantom) Standardized test images. Functions for generating standardized test examples such as `shepp_logan`.
 * [set](set) Sets of objects. Defines the abstract class `Set` and `LinearSpace` as well as some concrete implementations such as `RealNumbers`.
 * [solvers](solvers) Solution of equations and optimization. Contains both general solvers for problems of the form `A(x) = b` where `A` is an `Operator` as well as solvers of minimization problems. In addition, it defines the class `Functional` with several concrete implementations such as `L2Norm`.
-* [space](space) Concrete vector spaces. Contains concrete implementations of `LinearSpace`, including `NumpyFnBase` and `ProductSpace`.
+* [space](space) Concrete vector spaces. Contains concrete implementations of `LinearSpace`, including `NumpyTensorSpace` and `ProductSpace`.
 * [test](test) Tests for the ODL package. This contains automated tests for all other ODL functionality. In general, users should not be calling anything from this submoduel.
 * [tomo](tomo) Tomography. Defines the operator `RayTransform` as well as `Geometry` along with subclasses and utilities. Also defines problem dependent direct reconstruction such as `fbp_op`.
 * [trafos](trafos) Transformations between spaces. Defines `FourierTransform` and `WaveletTransform`.

odl/__init__.py

@@ -38,6 +38,9 @@
 except TypeError:
     pass
 
+# Set printing linewidth to 71 to allow method docstrings to not extend
+# beyond 79 characters (2 times indent of 4)
+np.set_printoptions(linewidth=71)
 
 # Propagate names defined in` __all__` of all "core" subpackages into
 # the top-level namespace

odl/contrib/fom/supervised.py

@@ -27,9 +27,9 @@ def mean_squared_error(data, ground_truth, mask=None, normalized=False):
 
     Parameters
     ----------
-    data : `FnBaseVector`
+    data : `Tensor`
         Input data to compare to the ground truth.
-    ground_truth : `FnBaseVector`
+    ground_truth : `Tensor`
         Reference to compare ``data`` to.
     mask : `array-like`, optional
         If given, ``data * mask`` is compared to ``ground_truth * mask``.
@@ -85,9 +85,9 @@ def mean_absolute_error(data, ground_truth, mask=None, normalized=False):
 
     Parameters
     ----------
-    data : `FnBaseVector`
+    data : `Tensor`
         Input data to compare to the ground truth.
-    ground_truth : `FnBaseVector`
+    ground_truth : `Tensor`
         Reference to compare ``data`` to.
     mask : `array-like`, optional
         If given, ``data * mask`` is compared to ``ground_truth * mask``.
@@ -137,9 +137,9 @@ def mean_value_difference(data, ground_truth, mask=None, normalized=False):
 
     Parameters
     ----------
-    data : `FnBaseVector`
+    data : `Tensor`
         Input data to compare to the ground truth.
-    ground_truth : `FnBaseVector`
+    ground_truth : `Tensor`
         Reference to compare ``data`` to.
     mask : `array-like`, optional
         If given, ``data * mask`` is compared to ``ground_truth * mask``.
@@ -198,9 +198,9 @@ def standard_deviation_difference(data, ground_truth, mask=None,
 
     Parameters
     ----------
-    data : `FnBaseVector`
+    data : `Tensor`
         Input data to compare to the ground truth.
-    ground_truth : `FnBaseVector`
+    ground_truth : `Tensor`
         Reference to compare ``data`` to.
     mask : `array-like`, optional
         If given, ``data * mask`` is compared to ``ground_truth * mask``.
@@ -268,9 +268,9 @@ def range_difference(data, ground_truth, mask=None, normalized=False):
 
     Parameters
     ----------
-    data : `FnBaseVector`
+    data : `Tensor`
         Input data to compare to the ground truth.
-    ground_truth : `FnBaseVector`
+    ground_truth : `Tensor`
         Reference to compare ``data`` to.
     mask : `array-like`, optional
         Binary mask or index array to define ROI in which FOM evaluation
@@ -336,9 +336,9 @@ def blurring(data, ground_truth, mask=None, normalized=False,
 
     Parameters
     ----------
-    data : `FnBaseVector`
+    data : `Tensor`
         Input data to compare to the ground truth.
-    ground_truth : `FnBaseVector`
+    ground_truth : `Tensor`
         Reference to compare ``data`` to.
     mask : `array-like`, optional
         Binary mask to define ROI in which FOM evaluation is performed.
@@ -406,9 +406,9 @@ def false_structures(data, ground_truth, mask=None, normalized=False,
 
     Parameters
     ----------
-    data : `FnBaseVector`
+    data : `Tensor`
         Input data to compare to the ground truth.
-    ground_truth : `FnBaseVector`
+    ground_truth : `Tensor`
         Reference to compare ``data`` to.
     mask : `array-like`, optional
         Binary mask to define ROI in which FOM evaluation is performed.
@@ -471,9 +471,9 @@ def ssim(data, ground_truth,
 
     Parameters
     ----------
-    data : `FnBaseVector`
+    data : `Tensor`
         Input data to compare to the ground truth.
-    ground_truth : `FnBaseVector`
+    ground_truth : `Tensor`
         Reference to compare ``data`` to.
     size : odd int
         Size in elements per axis of the Gaussian window that is used
@@ -536,9 +536,9 @@ def psnr(data, ground_truth, normalized=False):
 
     Parameters
     ----------
-    data : `FnBaseVector`
+    data : `Tensor`
         Input data to compare to the ground truth.
-    ground_truth : `FnBaseVector`
+    ground_truth : `Tensor`
         Reference to compare ``data`` to.
     normalized : bool
         If true, normalize ``data`` and ``ground_truth`` to have the

odl/contrib/tensorflow/README.md

@@ -8,7 +8,7 @@ This package contains ODL functionality related to [TensorFlow](https://www.tens
   This allows using tensorflow neural networks as operators in ODL.
 * `as_tensorflow_layer` in [layer.py](layer.py) wraps an ODL operator into a tensorflow layer.
   This allows using arbitrary ODL operators inside tensorflow neural networks.
-* `TensorflowSpace` in [space.py](space.py) is a `FnBase`-type space which uses tensorflow as a backend.
+* `TensorflowSpace` in [space.py](space.py) is a `TensorSpace`-type space which uses tensorflow as a backend.
 
 ## Example usage
 

odl/contrib/theano/layer.py

@@ -37,8 +37,8 @@ def __init__(self, operator):
         Parameters
         ----------
         operator : `Operator`
-            The operator that should be wrapped, must map `FnBase` spaces to
-            `FnBase` spaces.
+            The operator that should be wrapped, must map `TensorSpace`'s to
+            `TensorSpace`'s.
 
         Examples
         --------

odl/deform/linearized.py

@@ -8,14 +8,14 @@
 
 """Operators and functions for linearized deformation."""
 
-from __future__ import print_function, division, absolute_import
+from __future__ import absolute_import, division, print_function
+
 import numpy as np
 
-from odl.discr import DiscreteLp, Gradient, Divergence
+from odl.discr import DiscreteLp, Divergence, Gradient
 from odl.operator import Operator, PointwiseInner
 from odl.space import ProductSpace
-from odl.util import signature_string, indent
-
+from odl.util import repr_string, signature_string_parts
 
 __all__ = ('LinDeformFixedTempl', 'LinDeformFixedDisp', 'linear_deform')
 
@@ -146,21 +146,23 @@ def __init__(self, template, domain=None):
 
         >>> space = odl.uniform_discr(0, 1, 5, interp='nearest')
         >>> template = space.element([0, 0, 1, 0, 0])
-        >>> op = LinDeformFixedTempl(template)
+        >>> op = odl.deform.LinDeformFixedTempl(template)
         >>> disp_field = [[0, 0, 0, -0.2, 0]]
-        >>> print(op(disp_field))
-        [ 0.,  0.,  1.,  1.,  0.]
+        >>> op(disp_field)
+        uniform_discr(0.0, 1.0, 5).element([ 0.,  0.,  1.,  1.,  0.])
 
         The result depends on the chosen interpolation. With 'linear'
         interpolation and an offset equal to half the distance between two
         points, 0.1, one gets the mean of the values.
 
         >>> space = odl.uniform_discr(0, 1, 5, interp='linear')
         >>> template = space.element([0, 0, 1, 0, 0])
-        >>> op = LinDeformFixedTempl(template)
+        >>> op = odl.deform.LinDeformFixedTempl(template)
         >>> disp_field = [[0, 0, 0, -0.1, 0]]
-        >>> print(op(disp_field))
-        [ 0. ,  0. ,  1. ,  0.5,  0. ]
+        >>> op(disp_field)
+        uniform_discr(0.0, 1.0, 5, interp='linear').element(
+            [ 0. ,  0. ,  1. ,  0.5,  0. ]
+        )
         """
         space = getattr(template, 'space', None)
         if not isinstance(space, DiscreteLp):
@@ -231,11 +233,24 @@ def derivative(self, displacement):
         return PointwiseInner(self.domain, def_grad)
 
     def __repr__(self):
-        """Return ``repr(self)``."""
+        """Return ``repr(self)``.
+
+        Examples
+        --------
+        >>> space = odl.uniform_discr(0, 1, 5)
+        >>> template = space.element([0, 0, 1, 0, 0])
+        >>> op = odl.deform.LinDeformFixedTempl(template)
+        >>> op
+        LinDeformFixedTempl(
+            uniform_discr(0.0, 1.0, 5).element([ 0.,  0.,  1.,  0.,  0.])
+        )
+        """
         posargs = [self.template]
-        optargs = [('domain', self.domain, self.template.space)]
-        inner_str = signature_string(posargs, optargs, mod='!r', sep=',\n')
-        return '{}(\n{}\n)'.format(self.__class__.__name__, indent(inner_str))
+        optargs = [('domain', self.domain,
+                    self.template.space.real_space.tangent_bundle)]
+        inner_parts = signature_string_parts(posargs, optargs)
+        return repr_string(self.__class__.__name__, inner_parts,
+                           allow_mixed_seps=False)
 
 
 class LinDeformFixedDisp(Operator):
@@ -297,21 +312,23 @@ def __init__(self, displacement, templ_space=None):
 
         >>> space = odl.uniform_discr(0, 1, 5)
         >>> disp_field = space.tangent_bundle.element([[0, 0, 0, -0.2, 0]])
-        >>> op = LinDeformFixedDisp(disp_field)
+        >>> op = odl.deform.LinDeformFixedDisp(disp_field)
         >>> template = [0, 0, 1, 0, 0]
-        >>> print(op([0, 0, 1, 0, 0]))
-        [ 0.,  0.,  1.,  1.,  0.]
+        >>> op(template)
+        uniform_discr(0.0, 1.0, 5).element([ 0.,  0.,  1.,  1.,  0.])
 
         The result depends on the chosen interpolation. With 'linear'
         interpolation and an offset equal to half the distance between two
         points, 0.1, one gets the mean of the values.
 
         >>> space = odl.uniform_discr(0, 1, 5, interp='linear')
         >>> disp_field = space.tangent_bundle.element([[0, 0, 0, -0.1, 0]])
-        >>> op = LinDeformFixedDisp(disp_field)
+        >>> op = odl.deform.LinDeformFixedDisp(disp_field)
         >>> template = [0, 0, 1, 0, 0]
-        >>> print(op(template))
-        [ 0. ,  0. ,  1. ,  0.5,  0. ]
+        >>> op(template)
+        uniform_discr(0.0, 1.0, 5, interp='linear').element(
+            [ 0. ,  0. ,  1. ,  0.5,  0. ]
+        )
         """
         space = getattr(displacement, 'space', None)
         if not isinstance(space, ProductSpace):
@@ -374,11 +391,27 @@ def adjoint(self):
         return jacobian_det * self.inverse
 
     def __repr__(self):
-        """Return ``repr(self)``."""
+        """Return ``repr(self)``.
+
+        Examples
+        --------
+        >>> space = odl.uniform_discr(0, 1, 5)
+        >>> disp_field = space.tangent_bundle.element([[0, 0, 0, -0.2, 0]])
+        >>> op = odl.deform.LinDeformFixedDisp(disp_field)
+        >>> op
+        LinDeformFixedDisp(
+            ProductSpace(uniform_discr(0.0, 1.0, 5), 1).element(
+                [
+                  [ 0. ,  0. ,  0. , -0.2,  0. ]
+                ]
+            )
+        )
+        """
         posargs = [self.displacement]
         optargs = [('templ_space', self.domain, self.displacement.space[0])]
-        inner_str = signature_string(posargs, optargs, mod='!r', sep=',\n')
-        return '{}(\n{}\n)'.format(self.__class__.__name__, indent(inner_str))
+        inner_parts = signature_string_parts(posargs, optargs)
+        return repr_string(self.__class__.__name__, inner_parts,
+                           allow_mixed_seps=False)
 
 
 if __name__ == '__main__':