Hmm … that is strange. I checked this myself where e was the return value of map_overlap(), and it works very well. Perhaps you’re using an older version of dask?
The dask.dataframe module typically addresses different issues from those addressed by dask.array. And those issues that dask.array addresses can usually be done orders of magnitude faster by dask.array than by dask.dataframe. See for example this post and those before it. It takes a bit of experience, but I’m sure you’ll soon see that to be true for yourself. 