mvf 0.0.3

A package implementing a supervised learning model validation framework.

MVF stands for model validation framework. MVF is a pluggable ML/statistical modelling framework that allows for the easy comparison of models implemented in Python and R. Write simple wrapper classes for your models and compare their performance on a particular dataset.

Getting started

For full documentation of the project and instructions on how to get started, visit the documentation site.

Main features

For developers

Dependencies

R dependencies are managed using renv. Spin up a Python Virtual Environment

Python dependencies are specified by requirements.txt. This file is generated from requirements.in by running python3 -m piptools compile.

Git

This project operates using two Git branches

• dev
• main

All development work should be undertaken on the development branch. The dev branch should then be merged into the master branch to deploy a new version of the package.

CI/CD

This project uses GitLab CI/CD. There are currently three stages in the CI/CD pipeline

• test - Runs any CI/CD tests using pytest.
• build_deploy_package - Builds the Python package and deploys to PyPI.
• build_deploy_docs - Builds the documentation site and deploys to GitLab Pages.

The test stage runs on every commit. The build_deploy_package and build_deploy_docs stages only run on commits to the main branch. All CI/CD stages run in a Docker container. This project uses node:latest for the build_deploy_docs stage and a custom R/Python container for the remaining stages.

Docker

To update the container in the registry, navigate to the project root and run

sudo docker login registry.gitlab.com

Enter your GitLab username and password (only for members of the project). Then run

sudo docker build -t registry.gitlab.com/tomkimcta/model-validation-framework .
sudo docker push registry.gitlab.com/tomkimcta/model-validation-framework

PyPI

The version stored in the version file must be incremented for a deployment of the package to be successful.

Documentation

This project uses a static site generator called Docusaurus to create its documentation. The content for the documentation site is contained in documentation/docs/. Any updates to documentation can be verified in a development server by running npm i && npm start from the documentation/ directory.