Per-notebook environments

[grst]

it should be possible to define dependencies per notebook (conda envs?)

[grst]

@Zethson @ivirshup one issue is that binder only supports one environment.yaml per repo (and they don't want to support multiple ones: jupyterhub/binderhub#555).

A workaround they suggest is to !pip install packages at the beginning of notebooks as it's commonly done in colab.

What do we do? Is there any chance we can get away with a single environment for all notebooks?

[Zethson]

Not a big fan of having a list of pip installs in the beginning. I'd even argue that if our ecosystem is very compatible, a single environment should work out. It should work for all scverse core packages.

I wouldn't be surprised however if many of the scverse ecosystem packages would not fit into such a massive environment. Some of them pin some fundamental packages.

[grst]

If we were to ever use the galaxy jupyter, how would we define dependencies here?

I tried setting up a binder with a docker container that contains multiple conda envs, but it doesn't want to start for me:
https://github.com/grst/test-binder-conda-docker

[grst]

For now, going for a single environment.yaml, because we need something that works.

Eventually, we'll need to figure something out, but I still don't know what a solution that works with binder could look like (unless someone else makes the multi-env docker work)