[Feature request] Python bindings

[apcamargo]

Hi,

I've been using annembed for some time now and I'm really satisfied with how fast it is. However, the workflow from Python is very cumbersome at the moment. It would be great if there were Python bindings that allowed annembed to be called from within Python, using numpy arrays as input.

[jianshu93]

@jean-pierreBoth,

do you think we need to have it since it is really useful in a python environment, comparable to UMAP. @apcamargo We do have some comparison with UMPA in terms of visualization accuracy in the paper: https://www.biorxiv.org/content/10.1101/2024.01.28.577627v1.abstract

best,

Jianshu

[apcamargo]

I think this can be really useful both because of the advantages regarding visualization and because it is way faster than UMAP (and would require less dependencies too)

[apcamargo]

I can try to create a fork a develop Python library using PyO3. But it would be a couple of months before I have the time to start working on that.

[jianshu93]

hi @apcamargo,

Annembed paper was published here: https://academic.oup.com/nargab/article/6/4/lqae172/7928174, feel free to cite it if you find it useful. And we look forward to the python bindings.

Best,

Jianshu & Jean

[jianshu93]

Hi @apcamargo,

python bindings for annemeb is finally here: https://pypi.org/project/annembed-rs/

usage can be found in the readme of this repo.

Best,

Jianshu & Jean

[apcamargo]

Thank you, @jianshu93! This will help a lot!

Do you guys plan to make it possible to provide a numpy array as input? Just to have an API closer to what UMAP and scikit-learn provide. It could make things easier for the user, allowing them to bypass saving data that they generated in a Python code into a csv just to read again in annembed.

Just a suggestion though. The current implementation will already be really useful for me!

[jianshu93]

The CSV interface now is very fast for 11 million samples, IO in python is generally slower and both UMAP and scikit-learn cannot scale to millions of data points. Saving data in a TMP directory and load them back for small number of samples should be very fast. The output is in Numpy already for visualization et.al.. In any case, thanks for your suggestion. We may add this option in the future. --Jianshu & Jean