Cannot install modules inside Docker container with Jupiter notebook image

Question

I have got the following Docker file

FROM jupyter/scipy-notebook

COPY . ./work

RUN pip install -r ./work/requirements.txt

COPY setup.py /work

RUN pip install --upgrade pip && pip install -e .



COPY src /work/src

WORKDIR /work

and the following project structure:

almond_analysis:
    notebooks:
        data_exploration.ipynb
    src:
       __init__.py
       plots.py
    .gitignore
    docker-compose.yml
    Dockerfile
    README.md
    requirements.txt
    setup.py

Inside the data_exploration.ipynb notebook, I would like to import the module src with the command import src as test. However, despite me having typed RUN pip install -e . in the Dockerfile, I get the error ModuleNotFoundError: No module named 'src'. When I do to my container though, inside the work directory, and run pip install -e . the error goes away. I tried adding CMD pip install -e . at the end of the Dockerfile without success. I read also what was suggested in this post (more specifically to add the lines RUN apt-get update and RUN apt-get install -y --no-install-recommends but without success.

My setup.py file looks like this:

from setuptools import find_packages, setup
import sys
import os

sys.path.insert(0, os.path.abspath( os.path.dirname(__file__)))


requirements_path="requirements.txt"
with open(requirements_path) as requirements_file:
    requirements = requirements_file.read().splitlines()

setup(
    name="almond_analysis",
    version="0.0.1",
    description = "Almond analysis",
    long_description="Analysing yield with Python."
    author= "George N",
    packages=find_packages(),
    install_requires=requirements,
    classifiers=[
        "Programming Language :: Python :: 3"
    ],
    python_requires =">= 3.0.*",
)

Does someone have an idea on how to make the ModuleNotFoundError go away?

Do you need to "install" `src`? From the notebook, you should already be able to do `from src import plots`. Without seeing your setup file, we don't know what the real module name should be — OneCricketeer, Jun 22 '22 at 16:00
Okay, so looks like you should be trying to import `almond_analysis`, not `src` since that is the name of the module, but like I said, you should also be able to import directly from the folder/package without "installing" anything else. — OneCricketeer, Jun 22 '22 at 20:04
I cannot import directly from the folder/package right after I have typed `docker compose up` . If I go into `data_exploration.ipynb` and type `import src as test` or `from src import plots`, I get the error `ModuleNotFoundError: No module named 'src'`. However, if I go inside my container (inside the `work/` directory) and type `pip install -e .`, both these errors go away. I want not to have to go in the container and type `pip install -e .`, but instead for this command to be run in the `RUN pip install --upgrade pip && pip install -e .` line in my Dockerfile. — JohnK, Jun 23 '22 at 11:16
The module you are installing is called `almond_analysis`, not `src`, though. — OneCricketeer, Jun 23 '22 at 17:05

OneCricketeer · Accepted Answer · 2022-06-27T16:33:28.637

Mount your code under the work folder, then you don't need to install anything, or really need a Dockerfile unless you have other dependencies.

On host, create folder for notebooks...

notebooks/src/plots.py

def foobar():
    print('Hello from plots')

In compose, mount it under the working directory /home/jovyan/work

notebook:
    image: jupyter/scipy-notebook
    ...
    volumes:
      - ./notebooks:/home/jovyan/work

Load up the Jupyter environment, and create a notebook in the work folder, and then import the module.

from src import plots
plots.foobar()  # Hello from plots

This is the same workflow you'd do on your host if you weren't using Docker or Jupyter, so neither of those are really the problem.

Cannot install modules inside Docker container with Jupiter notebook image

1 Answers1