Installing Python Packages Using Jupyter Notebooks with Bridges OnDemand

Q&A
, , , ,
1

I’m trying to install pytorch and pytesseract on PSC Bridges and I want to use that on a compute node, running in Jupyter notebook via OnDemand. I’ve installed both packages using the login node and pip install --user pytorch pytesseract and that installs correctly, but I can’t get it to load the module when I run the Jupyter notebook with OnDemand. We also looked at this blog post that seems to point to part of an answer: https://jakevdp.github.io/blog/2017/12/05/installing-python-packages-from-jupyter/

3

Based on wirawan0’s comment I see the problem now.

Bridges jupyterhub uses a conda environment: /opt/packages/anaconda/anaconda3-5.1.0
If you want to access that environment from the login node you could activate it:

source /opt/packages/anaconda/anaconda3-5.1.0/bin/activate
pip install torch --user

It’s probably more consistent to launch juptyerhub, open a terminal, and then pip install torch --user because that will always use the correct conda environment.

Thanks,
Mark

PS You can see where jupyterhub is loaded from using:

import sys
print (sys.prefix)

4

My immediate question is: does Jupyter notebook launched by Bridges’ Open OnDemand even support virtual environment? If it does not, then it is not straightforward to “fudge” Jupyter to use virtual environment.

2

Does the package work correctly in a regular Python environment (not within Jupyter notebook)?
If it does not - what kind of error do you get and when? Does the error occur during import command, or when you use a function from this package?