Baker Project

About

This project is named Baker because it provides a Docker image for Bayesian analysis. In particular, it enables

PyMC3 with Jupyter and miniconda3
Stan using R through rstan and other related packages
BUGS, in particular OpenBUGS, using R2OpenBUGS
JAGS using rjags

The website for this project is available at here. The project’s goal is to help

Students to not need to install packages and complicated dependencies on their personal devices
Professors, Instructors, and TAs to relieve from being system admins for all their students
Practitioners to quickly spin up an environment to test out something interesting they find on the internet and to share their work with the world in an easily reproducible manner

If you find this project helpful, please consider staring it and sharing the words. If you have any comments, concerns, or questions, please feel free to raise an issue.

Motivation

I was first introduced to WinBUGS in an undergraduate Bayesian analysis course. Not knowing much modern computing technologies at the time, I had to visit the lab to use their Windows machines which have the software installed. Then, a couple years flew by, I took another Bayesian class in grad school which also required OpenBUGS. To make things worse, iOS dropped support for 32 bit programs in the middle of the semester after I spent significant amount to time configuring wine, OpenBUGS, and R2OpenBUGS on my MacBook. I know now is the time.

Be the change you wish to see. —- Mahatma Gandhi

Usage

Some basic (very little) general knowledge of Docker is needed. If one is interested in learning more, this tutorial is a great source. We will present below a few common use cases.

The Dockerfile of this project is based on rocker/verse:latest, therefore sharing many similar usage. For example, to set up a rstudio server on local host, run

docker run \
    -d \
    --rm \
    -p 8787:8787 \
    -e USER=<username> \
    -e PASSWORD=<password> \
    --name <container_name> \
    haoencui/baker

Then, visit localhost:8787 in your browser with specified <username> and <password>.

-d: to start a container in detached mode;
--rm: automatically remove the container when it exits;
-p: publish container port to host, hostPort:containerPort. One can pass multiple -p options. 8787 is used by rstudio server;
-e: set environment variable, i.e. username and password for rstudio server. PASSWORD is required and cannot be "rstudio". If USER is omitted, then "rstudio" is used as default; and
--name: name of container for identification, can be omitted.

The --volume option (or -v in short) is also convenient to mount volume from local host (e.g. your laptop). For example, adding

--volume $(pwd):/home/<username>/WORKSPACE

to the aforementioned docker run command will connect ~/WORKSPACE inside the Docker container with current working directory (pwd) on your local host, i.e. the file system under the directory is shared and hence modification will be reflected in real time. For additional details on running rstudio server, please read rocker wiki page.

To set up a jupyter lab on local host, run

# WARNING: THIS PROBABLY WON'T WORK AND IS RUN AS ROOT WITHOUT AUTH
docker run \
    -d \
    --rm \
    -p 9999:9999 \
    --volume $(pwd):/WORKSPACE \
    haoencui/baker \
    /bin/bash -c \
    "jupyter notebook --allow-root --ip=0.0.0.0 --port=9999 --no-browser --NotebookApp.token='' --NotebookApp.password=''"

Then, visit localhost:9999 in your browser without the need to sign in.

If one needs to peek into the Docker container while it’s running, use

docker ps

to find the <container-id> and run as root user via

docker exec -it <container-id> bash

and this will start a bash shell running inside the Docker container as identified by its ID.

For one-off scripts, one can execute them via

docker run --rm haoencui/baker /bin/bash -c "<some-command>"

For example, <some-command> can be echo $HOME && echo $USER or Rscript file.R. Don’t forget the --volume option if you want to access or modify files on local host. But note that this will run as the root user so don’t do something too crazy if you are also exposing your laptop’s file system.

Last but not least, don’t forget to shutdown the containers using

docker stop <container-id>

and remove stopped containers, if you’d like

docker container prune

Contribution

Users wanted: Please try out this project and let me know your thoughts. There are a few examples on the project website if you don’t have a Bayesian model handy.

I hacked together this project over a weekend before Christmas. This is also the first Dcokerfile I’ve ever built, so there are probably a lot of problems and room for improvements. Please feel free to raise an issue and / or directly contribute. Your help is greatly appreciated.

Acknowledgement

This project is greatly inspire by the rocker project, [@jrnold's `rstan` `Dockerfile`](https://hub.docker.com/r/jrnold/rstan/dockerfile), and [@jonzelner's `rstan` `Dockerfile`](https://hub.docker.com/r/jonzelner/rstan/dockerfile).

Special thanks to [@jsal13](https://github.com/jsal13) and [@patrick-boueri](https://github.com/patrick-boueri) for helping me debug my messy docker build logs.