Use numpy histogram in scatter_density

[quentinblampey]

Hello,

These are the changes related to #64, i.e. about the usage of np.histogram2d instead of datashader in scatter_density: the performances are the same, as shown below.

I kept the same function signature, except for cmap where str | Colormap is supported, but not List yet (do we often need to use a list of colors as a cmap?)

Since datashader is not used in pytometry anymore, I removed it from the dependencies (as long as it's not needed for another plot), and I also moved nbproject in the test dependencies.

I also added a CHANGELOG.md file, I think it's a good practice to have this to keep track of the changes per version, what do you think @mbuttner, @grst?

With datashader

%%time
pm.pl.scatter_density(adata, x="CD4", y="CD8")

CPU times: user 494 ms, sys: 7.36 ms, total: 501 ms
Wall time: 210 ms

With numpy histogram2d

%%time
pm.pl.scatter_density(adata, x="CD4", y="CD8")

CPU times: user 403 ms, sys: 7.67 ms, total: 411 ms
Wall time: 203 ms

[github-actions]

🚀 Deployed on https://664753f21ee1d0c61500dd34--pytometry.netlify.app

[grst]

LGTM, just one comment RE changelog.

[mbuttner]

Looks good to me, too!

[Zethson]

To be honest, I'm not sure why you need nbproject at all. For tests, you only need nbproject_test.

It's used once in the code:

from nbproject._logger import logger (https://github.com/scverse/pytometry/blob/main/tests/test_notebooks.py#L4) and this can easily be replaced with just a print or Python logger statement

[Zethson]

This is a general point, but I don't see the point of typing docstrings when you're using typehints in the function stub (as you should!). It's annoying to keep both in sync. Sphinx picks up the types from the function header anyways.

[quentinblampey]

I totally agree, I'll open a separate issue!