mvf

MVF is a pluggable ML/statistical modelling framework that allows for the easy comparison of regression models implemented in Python and R. Write simple wrapper classes for your models and compare their performance on a particular dataset. The framework executes incrementally by default, meaning you can easily add a new model, without refitting existing models.

Pluggable - Quickly add a new model. Easily switch-out a dataset.

Auditable - All results are archived. All processes are transparent.

Reproducible - All non-stochastic processes are reproducible. Stochastic processes are repeatable.

MVF seeks to bridge the gap between R&D and integration with production applications.

Getting started

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

Main features

• Automates the supervised ML workflow with simple configuration.
• R and Python models can be plugged in easily.

For developers

Dependencies

You need Python>=3.9 and R>=4.0.

Additionally, you must have a working installation of the R6, IRkernel and arrow R packages to leverage the R/Python interoperability.

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. Current jobs:

• test_dev - Runs all tests except functional tests using pytest. Runs on commits to dev branch.
• test_main - Runs all tests using pytest. Runs on commits to main branch.
• build_deploy_package - Builds the Python package and deploys to PyPI. Runs on commits to main branch.
• build_deploy_docs - Builds the documentation site and deploys to GitLab Pages. Runs on commits to 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 specified by the Dockerfile 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.