Need to seek help from the almighty community again. I am currently setting up my python environment with a module like this:
{ config, pkgs, lib, ... }:
let cfg = config.vital.dev.python;
batteryOptions = {
options = {
machineLearning = lib.mkEnableOption ''
Include the machine learning libraries, e.g. pytorch and lightbm.
'';
# This is set to true.
jupyterhub = lib.mkEnableOption ''
Include jupyterhub.
'';
};
};
in {
options.vital.dev.python = {
batteries = lib.mkOption {
description = "Turn on batteries, which represents groups of python packages.";
type = lib.types.submodule batteryOptions;
};
};
config = {
environment.systemPackages = [
(let pythonWithBatteries = pkgs.python3.withPackages (
pythonPackages: with pythonPackages;
let the-torchvision = torchvision.override {
pytorch = pytorchWithCuda;
};
in [ numpy pandas matplotlib jupyterlab jupyterlab plotly dash tqdm ]
++ (lib.optionals cfg.batteries.machineLearning [ lightgbm pytorchWithCuda the-torchvision ])
++ (lib.optionals cfg.batteries.jupyterhub [ jupyterhub ])
);
in pythonWithBatteries)
];
};
}
Starting jupyterhub with default jupyterhub config leads to using jupyter notebook. What is the best way to achieve having a jupyterhub instance that runs jupyter lab server?
The only relevant issue I found is JupyterHub doesn't see JupyterLab · Issue #58662 · NixOS/nixpkgs · GitHub. I am wondering how I can override the propagatedBuildInputs? The dumb method I can think of is to copy the whole package definition locally, modify it and import it in my overlay to shadow the original jupyterhub defined in <nixpkgs>. Is there a better way to do so?
I use the thing you may see below ( it is obtained and combined from posts on this channel).
I left some comments in there if you want to play & try. And sorry for not directly answering your question (I hope this helps).
Random comments:
there is examples on how to add your own packages (I just marked the path with dots).
(one such example package definition for is even more below)
I think I tested with several kernels but not sure (don’t need it, this is older than three days so don’t remember any more).
it uses 3.7 because those tensorflow things didn’t yet work on 3.8 (haven’t checked the situation lately)
shows how to add extensions
shows how to add packages
shows how to add other programs / packages
the shellHook at the end is really bad but does it’s job
unfortunately, I don’t remember, how I was able to try to find out the correct commands
so, somebody may answer or tell me, how I did it I’d prefer someone from the marketing thread: if possible, this is something that I’d like to be easily available information on how to use nix
so, why I said this bad (I mean ugly), you see, the scripts calls jupyterlab build twice
the reason is that the first one fails because of missing directories but creates them
and then the second run succeeds by being now able to use those directories
as a last thing, it starts the jupyter lab
you may leave it out, if you want start it manually (just give that command on shell then)
I hope that this helps even though this doesn’t answer your question directly.
Hi @gspia, thanks a lot for sharing your configurations. I am going to try it tonight!
Great to learn how you add jupyter lab extensions. Really appreciate it!
I’d prefer someone from the marketing thread: if possible, this is something that I’d like to be easily available information on how to use nix