* Use 'masked' argument to return masked array based on nodatavals
* Needs more verification and unit testing to ensure it works, and
this is open to any other addition/removal of features as this is
a new high-level function.
Other changes in this commit:
* Check 'window' parameter structure, and raise ValueError if not correct
* Cython 0.21dev has difficulty compiling e.g. `set([0]).pop()`, so
modify how 'meta' property works, for now.
* Allocate Numpy arrays with 'empty' rather than 'zeros', since they are
expected to be filled with values from raster.
* setup.py: make an executable, 'find_packages' not used, and avoid
cythonizing new files on './setup.py clean' command
Both ndarray to ndarray and inter-dataset reprojections are
possible. TODO:
- Add the finer warping options.
- Use valid data masks when warping.
- Test data types other than uint8.
- Make warping of big datasets chunky. Though we do have the
option of rasterio dataset windows.
Lots of improvements to be made. Much cargo, very culting. I've
still a lot to learn in the intersection of Cython, C++, and
GDAL's warping API.
Closes#12.
GeoJSON-ish features are excessive, so now shapes() simply yields
a geom, image_value pair. This let me delete a bunch of Cython code.
It's easy enough to make features out of these if needed.
