Chris Barnes

This looks like it would be a good fit: https://pypi.org/project/shared-ndarray2/ , but it's py3.8+ only. Those memory drops might be the garbage collection running, I guess?

You could also avoid the copies all being done at the end by writing them in to the dataframe as they come out: ```python scores = pd.DataFrame(np.zeros((len(query_dps), len(target_dps)), dtype=this.dtype), index=query_dps.id,...

I've voiced my opinions on bumping the minimum version :sweat_smile: It doesn't break existing versions of navis, which people are welcome to keep using on their old python versions (this...

Just raised a PR on reticulate https://github.com/rstudio/reticulate/pull/1185 which (I think) explicitly runs tests against python 3.8, 3.9, and 3.10; passes for me https://github.com/clbarnes/reticulate/actions/runs/2090867132 !

Seems to be whatever python happens to be lying around in the CI environment. Fortunately a certain subset of the python ABI doesn't change very quickly so I wasn't expecting...

Reticulate now explicitly tests against 3.8-10!

Is there ever a situation where you'd want to compare two dotprops of different k? If not, you could store k on the dotprops object and potentially check it during...

It's been thought of, but is tricky due to the underlying compiled libraries. I think there might be some dynamic linking which makes building wheels hard. [Pyn5](https://pypi.org/project/pyn5/) is a very...

I forget, do the zarr/N5 specs support dataset reshaping? They'd handle the edge chunks quite differently. Otherwise, on the application side you could set aside a metadata key like "followed_by"...

I think @aschampion designed the [benchmarks](https://github.com/aschampion/rust-n5/blob/master/benches/parallel_write.rs) for rust-n5 to match some benchmarks being done on the [java reference implementation](https://github.com/saalfeldlab/n5/) of N5, ~~although I don't know where the java benchmarks are,~~...