36

I have used jupyter notebook for data analysis for quite sometime. I would like to develop a module in my jupyter notebook directory and be able to import that new module into notebooks. My jupyter notebook file directory can be represented as follows;

Jupyter notebooks\

    notebook1.ipynb

    new_module\
        __init__.py
        newfunction.py

    currentnotebooks\
        notebook2.ipynb

When use import new_module in notebook1.ipynb it works however when I try the same command in notebook2.ipynb I get the following ImportError: No module named 'new_module'. The two obvious solutions are A) move new_module into the currentnotebooks directory or B) move notebook2.ipynb up to the same level as new_module. I don't want to mess around with the file structure at all. Is this possible?

Thomas K
  • 39,200
  • 7
  • 84
  • 86
James Draper
  • 5,110
  • 5
  • 40
  • 59
  • If you put `new_module` into one of the directories listed in `sys.path`, it will be importable from anywhere. – Thomas K Sep 03 '16 at 09:31

1 Answers1

52

You need to make sure that the parent directory of new_module is on your python path. For a notebook that is one level below new_module, this code will do the trick:

import os
import sys
nb_dir = os.path.split(os.getcwd())[0]
if nb_dir not in sys.path:
    sys.path.append(nb_dir)

If you're further down in the directory hierarchy, you will need to adjust the way nb_dir is set, but that's all. You should not run this code for a notebook in Jupyter notebooks, since it would add the parent of that directory to the python path, which is probably undesirable.

The reason the import works for notebook1 is that sys.path contains '' (the empty string), which is aways the current directory of the running interpreter (kernel, in this case). A google search for explain python path turns up several good explanations of how python uses PYTHONPATH (aka sys.path).

cco
  • 5,873
  • 1
  • 16
  • 21
  • 1
    So I always have to do your suggested boilerplate code before I can access other modules even if they are just one folder above? Is there no other proper way to achieve this, e.g. via relative imports? (though I must admit that I am still not sure about how relative imports work) – LFish Jul 23 '17 at 16:05
  • 1
    More or less. The exception would be if the parent directory was a module as well (and so had an`__init__.py` file), but that seems both strange and unlikely to me. Putting the notebook directory in `PYTHONPATH` would avoid the need to add it to sys.path this way. – cco Jul 23 '17 at 22:57