Final Project Repository Template

This is the final project repository template for Machine Learning with Probabilistic Programming.

Duplicating your own copy of this repository

Please follow these instructions to make a copy of this repository and push it to your own GitHub account.

Make sure to create a new repository on your own GitHub account before starting this process.

Final Project Notebook

We have included a example of a Jupyter notebook under /notebook-example/example.ipynb. This shows how to use markdown along with LaTeX to create section headings and typeset math.

Your final project notebook should go under /final-project/final-notebook.ipynb. This notebook will be your final report. We must be able to run it in a reasonable amount of time. (If your project involves a massive dataset, please contact me.)

Your final report should be 8 pages long. Since it is hard to translate between a Jupyter notebook and page numbers, we've come up with the following metric:

the Markdown export of your notebook should be approximately 1500 words.

To compute this, save your Jupyter notebook as a Markdown file by going to

File > Download as > Markdown (.md)

and then counting the words

wc -w final-notebook.md

Since this includes your code as well, we encourage you to develop separate python scripts and include these in your final notebook. My recommendation is that you only do basic data loading, manipulation, and plotting within Jupyter; do all of the heavy lifting in separate Python files. (Note our strict guidelines on coding style below.)

Structure

Your notebook should follow the basic structure described in the project proposal template. Make sure to clearly indicate section headings and to present a clear narrative structure. Every subsection of your report should correspond to a particular step of Box's loop. Feel free to include images; you can embed them in markdown cells.

Development

Use Python 3.6+. (I use Python 3.6.5).

Configure a virtual environment. Follow the documentation here. (I like to use virtualenvwrapper.)

Once you activate the virtual environment, use pip to install a variety of packages.

(venv)$ pip install -r requirements.txt

This should install Edward, along with Jupyter and other useful libraries. You should see a message at the end that resembles something like

Successfully installed appnope-0.1.0 bleach-1.5.0 ...

Additional dependencies

If you introduce any new dependencies to your final project, you MUST update requirements.txt with pinned versioning.

If you plan to introduce any non-pip-installable (e.g. Stan) dependencies to your final project, you MUST provide a Dockerfile. (Please contact me before you do so.)

Git stuff

There is a comprehensive .gitignore file in this repository. This should prevent you from committing any unnecessary files. Please edit it as needed and do not commit any large files to the repository. (Especially huge datasets.)

Code styling

Any additional code you write must pass flake8 linting. See this blog post for details.

The first thing we will do after cloning your repository is:

(venv)$ pytest --flake8

If your repository fails any checks, we will deduct 20% from your final project grade.