There are many ways to make workflows reproducible. This repo serves to demonstrate a few options.
Mybinder is a free online service that builds a docker container with a jupyter server from a github repository. It looks for a config file in binder, which is used to build a conda environment with the necessary dependencies to run the notebook. You can go to mybinder and enter the URL of this repo there, or alternatively click the button below to start building the environment and serve up the notebook.
Since the data file is not included in the Github repository, if I delete the file on the file server, the notebook in the mybinder environment will cease to work.
In order to ensure everything is bundled together, we build a docker image. This bundles a minimalistic operating system, the necessary packages and software, and the data all together in one file that can be run as a docker container. To build such an image, the necessary configuration files Dockerfile and .dockerignore are included in this repo. To build the docker image yourself for this repo, you must:
$ git clone https://github.com/din14970/example_reproducible_workflow
$ cd example_reproducible_workflow
$ mkdir Data
$ wget -O "Data/example.emd" "https://owncloud.gwdg.de/index.php/s/wBifnrx7RpsP5nL/download"
$ docker build -t yourname/example_notebook .
or do equivalent steps manually. This presumes you are on a Unix based system, that the data file remains available, and that you have Docker CLI installed.
You can then run the image with
$ docker run -p 7000:8888 yourname/example_notebook
You can then navigate to http://localhost:7000 and run the notebook.
In principle I can also upload this image to docker hub with
$ docker push yourname/example_notebook
And this can be downloaded with
$ docker pull yourname/example_notebook
Alternatively, the entire image can be condensed to a tar archive
$ docker save -o example_notebook.tar yourname/example_notebook
which can then be uploaded anywhere, like on Github releases. This tar file could then be loaded as an image with
$ docker load example_notebook.tar
It is only such an image that is truly reproducible! However I haven't done this yet because I didn't want to upload 3.5 Gb from my home network.