An opinionated template repo for data science / ML pipelines in python

Read more about it in this blog post

• Modularity
  • Separate directories for data, notebooks, base package, and tests.
  • Different workflow components are separated from each other and from workflow management (i.e. Pipeline), to allow flexible recomposition.
  • Configurations are further separated via config files.
• Readability
  • Pipeline manager enables low-code notebooks designed for data exploration and workflow management.
  • Code formatting / linting / testing is enforced
  • Docstrings and type-hinting in function definitions
• Repeatability
  • Dependencies / environment managed through code (i.e. setup.sh), including installation of the base package.
  • Workflows instantiated through low-code scripts or notebooks, which could further be managed by a dedicated WMS.
• Robustness
  • Unit tests and code coverage requirements strictly enforced through
    • pre-commit hooks
    • CI pipeline through github actions
    • branch management.