The goal of this repository is to act as a template demonstrating a full end to end MLOps style R project.
The toy example is training a predictive model on a csv dataset that predicts water temperature throughout the year for Sydney beaches. This project addresses how you might want to structure this project to run in a ‘Production’ context.
I go into more details in my workshop Data Science Workflows in R An introduction to deploying production quality R code.
This includes:
- Data import and cleaning with the tidyverse
- Data documentation
- Documented R functions for processing and model training using roxygen2
- R package structure
- Model training and inference pipeline using quarto
- Unit tests with testthat
- Monitoring dashboard in interactive rmarkdown and flexdashboard
- Model card for documenting model ethics, decision and recommended use cases
- Prediction API endpoint creation using plumber
- Model versioning and metadata storage using vetiver and pins
- Fully containerized deployment using Docker
- Dependency managment using renv
- Project website and documentation using pkgdown
- Continuous Integration (CI) pipeline automated using Github Actions
Installation
You can install the development version of beachwatch from GitHub with:
# install.packages("pak")
pak::pak("deanmarchiori/r-model-template")Then run devtools::install()
The model pipeline can be stepped through in model_notebook.qmd
Example
This is a basic example of using the deployed outputs from this project.
Clone the repository:
git clone https://github.com/deanmarchiori/r-model-template.gitTo build your docker image:
docker build -t beach .To run it:
docker run -p 8080:8080 beach
The endpoint should now be live at http://127.0.0.1:8080