0.1.8
 - Updated documentation
 - Cleaned up examples
0.1.7
 - Move documentation from GitHub to readthedocs.io
 - Add universal wheel on PyPI
 - Update tests on travis with new versions of NumPy
0.1.6
 - Bugfixes:
  - size of reconstruction volume in z too large for cases where
    the y-size is larger than the x-size of the sinogram images
  - `backpropagate_3d_tilted` used wrong shape of projections
  - 3D backpropagation methods did not use power-of-two padding size
0.1.5
 - Code optimization (speed, memory) with numexpr
 - New keyword argument `save_memory` for 3D reconstruction
  on machines with limited memory
 - New keyword argument `copy` for 3D reconstruction to protect
  input sinogram data.
0.1.4
 - The exponential term containing the distance between center
  of rotation and detector `lD` is now multiplied with
  the factor `M-1` instead of `M`. This is necessary,
  because usually the scattered wave is normalized in both
  amplitude and phase (`u_0`) and not only amplitude `a_0`
 - Allow angles of shape (A,1) in `backpropagate_3d_tilted` 
 - Set default value lD=0 for all reconstruction algorithms
 - Improvement of documentation
0.1.3
 - Fixes for `backpropagate_3d_tilted` when `angles` are
  points on the unit sphere:
  - Make sure each point is normalized
  - Correctly rotate each point w.r.t. `tilted_axis`
0.1.2
 - Added reconstruction algorithm for tilted axis of rotation
0.1.1
 - Support NumPy 1.10.
 - Allow to weight backpropagation using keyword `weight_angles`
 - Bugfix: backpropagate_3d with keyword `onlyreal=True` did not work
 - Bugfix: sum_2d did not return correctly shaped array
 - Code coverage is now 90%
 - Added more examples to the documentation
