Instead of a fix, I dumped the whole rolling mean calculation and use scipy.ndimage.filters.uniform_filter1d instead which is faster and better (it handles edges well without just copying the neighbouring values). Besides the edge cases, it is fully backwards compatible (I ensured float computation even though you could argue that int would do the job too).
Fix #85 (see bug description there)
Instead of a fix, I dumped the whole rolling mean calculation and use
scipy.ndimage.filters.uniform_filter1d
instead which is faster and better (it handles edges well without just copying the neighbouring values). Besides the edge cases, it is fully backwards compatible (I ensured float computation even though you could argue that int would do the job too).Considerations regarding efficiency
Function Doc