Model Development Template

This is a git repo template for creating new ML models within Jupyter Notebooks and/or VS Code.

Features

• Development container oriented
• Jupyter Notebook oriented
• A basic Python package structure is included for migrating reusable code out of notebooks

Delimitations

• This is not for automated model training. The design focus is on interactive development of new algorithms.
• The idea is not to use the latest and coolest technologies, but rather to keep things fairly standard and boring such that the focus remains on modelling.

Usage

After cloning this repo, either use your terminal or Visual Studio Code (recommended).

Visual Studio Code

Install the Remote Containers extension and open this repo as your workspace. See .devcontainer for details.

Terminal

1. (optional) Bootstrap a developer toolchain via Docker by make shell (or manually install Python and direnv).
2. With repo root as working directory, create a Python virtual environment by direnv allow.
3. Install current Python dependencies with make install (reset by make clean).
4. Start a Jupyter Notebook server with jupyter notebook and browse to https://localhost:8888

Tip: It's also possible to execute notebooks on the command line with jupyter execute (as an alternative to "Restart and Run All") or ipython -i my_notebook.ipynb to halt on errors and drop into an IPython session for interactive debugging.

FAQ

docker compose build

If you get docker: 'compose' is not a docker command. when trying make shell or any of the Docker targets, make sure you've migrated to Compose V2 and try again.