A tool that supports one-button reproducible workflows with the Jupyter Notebook and Scons. Note: this currently only supports Python kernels.
UPDATE: Scons >= 3.0.0 now supports Python 3 so Python 2 isn't needed anymore! Now Nbflow and Scons are both Python 2 and 3 compatible so you can choose whichever you want.
The actual version of Scons (3.0.1) currently supports Python >= 3.5, which is the default in Ubuntu 16.04
To install, run:
Linux:
pip3 install git+git://github.com/jhamrick/nbflow.git
Windows:
pip install git+git://github.com/jhamrick/nbflow.git
For a complete example of how to use nbflow, check out the example in this repository.
You can now you Binder to check the example online:
- Entre in Binder here or through the badge above
- Open a terminal
- Run
cd nbflow/example
- Run
scons
- Check the results in the
results
directory
Optionally you can modify the notebook in this online environment and check how the results change.
For each notebook that you want executed, you MUST include two special variables in the first code cell:
__depends__
-- a list of relative paths to files that the notebook depends on__dest__
-- either a relative path, or list of relative paths, to files that the notebook produces
For example, the first cell in one of the example notebooks is:
__depends__ = ["../results/data.json"]
__dest__ = "../results/stats.json"
You need a SConstruct
file in the root of you analysis directory. In this
SConstruct
file you will need to import nbflow and use it to setup your scons
environment, e.g.:
import os
from nbflow.scons import setup
env = Environment(ENV=os.environ)
setup(env, ["analyses"])
The second argument of the setup
command takes a list of folder names that
contain analysis notebooks.
Once you have setup your analysis notebooks and your SConstruct
file, you can
run your notebooks by just running the scons
command from the root of your
analysis directory.