jakirkham
commented
7 years ago Some of these operations repeat until the result does not change further (e.g. binary_propagation). This would require while-loops in Dask, which certainly don't exist as they would introduce cycles. We can certainly consider implementing these sorts of things. However they are going to be a bit trickier. Perhaps one way to go about them would be to persist results at the end of each run and compute some metric determine if convergence has occurred. To avoid having the computation start right away, we can investigate wrapping up this while-loop style function inside of a Dask Delayed object. As those objects can easily be converted back to Dask Arrays, this should work reasonably well.
The list of functions below are what that this library intends to support with Dask Arrays. These all come from SciPy. In particular they come from
scipy.ndimage.morphology. The intent is to emulate their behavior with Dask Arrays as wrapping them is not likely to work. Will also include tests to verify these retain the behavior that the SciPy functions ordinarily have.binary_fill_holes(This would require awhile-loop in Dask to work.)binary_propagation(This would require awhile-loop in Dask to work.)generate_binary_structure(generates very small arrays for use with ops)iterate_structure(works with very small arrays)