search

brown-ccv / jupyterhub-docker-images

GNU General Public License v3.0
0 stars 1 forks source link

feat: add econ 2070 #11

Closed fernandogelin closed 4 years ago

fernandogelin commented 4 years ago

closes #16 closes #15

fernandogelin commented 4 years ago

Econ wants Julia as well. You probably need to reach out to get list of Julia packages though but you can set up few basics like DataFrames to start with

@mirestrepo where does it say that they want Julia? The ticket only says SciPy.

fernandogelin commented 4 years ago

reached out to Prof Koby. Waiting for feedback on Julia and other packages.

fernandogelin commented 4 years ago

@mirestrepo @mcmcgrath13

I believe setting the JULIA_PROJECT env is working if the packages get installed and compiled from the REPL. The Julia Kernels have --project=@. so the JULIA_PROJECT gets overwritten. To overcome that, I created a new kernel (Julia User), with --project=./julia/environments/v1.5.

Now, if an user installs a package, the installed packages/compilation seem to persist if we restart the server. However, the kernels won't connect anymore if the Project.toml in JULIA_PROJECT doesn't have IJulia. The kernels start fine if there's no Project.toml in the JULIA_PROJECT folder.

Any ideas would be appreciated.

fernandogelin commented 4 years ago

@mirestrepo @mcmcgrath13

I believe setting the JULIA_PROJECT env is working if the packages get installed and compiled from the REPL. The Julia Kernels have --project=@. so the JULIA_PROJECT gets overwritten. To overcome that, I created a new kernel (Julia User), with --project=./julia/environments/v1.5.

Now, if an user installs a package, the installed packages/compilation seem to persist if we restart the server. However, the kernels won't connect anymore if the Project.toml in JULIA_PROJECT doesn't have IJulia. The kernels start fine if there's no Project.toml in the JULIA_PROJECT folder.

Any ideas would be appreciated.

maybe a post-start hook? that installs IJulia in the JULIA_PROJECT?

fernandogelin commented 4 years ago

I kept the two Julia kernels until we figure this out, but ideally we would have just one with --project=./julia/environments/v1.5 that replaces the default.

mirestrepo commented 4 years ago

Actually,thinking more about it, lets only have the environment at /julia/environments/v1.5.... with the DEPOT_PATH set to home, recompilation works. Adding new packages won't be remembered, but that is the same behavior in the python side. If an user wants to install packages, they need to start a new environment in their home. And we just need to have that documented so they also now how to activate that in the notebook (They can simply do and ]activate path_to_env)

mirestrepo commented 4 years ago

So don't set the JULIA_PROJECT ENV, leave the default environment active and the kernel pointing to that

fernandogelin commented 4 years ago

@mirestrepo Can you review? I think it is working as intended. We will need to add instructions for users to create an env and activate before installing packages.