Abstract

Building Pipelines and Deploying Models with targets, tidymodels, and vetiver in R

This talk introduces a practical workflow for developing and deploying machine learning models using the targets and tidymodels frameworks in R. Our approach brings structure and reproducibility to the entire modelling process, from data preprocessing and feature engineering to model tuning and evaluation. We show how workflowsets simplifies model comparison, stacksenables ensemble learning, probably facilitates model-agnostic uncertainty quantification, and DALEX supports model interpretation and explainability.

To integrate modelling with production, we further cover essential MLOps concepts using vetiver for deployment, plumber for building web APIs, and Shiny for interactive visualisation. We also demonstrate how model cards can document model performance, assumptions, and intended use to support transparent and responsible deployment. Docker and Kubernetes are used for scalable, containerised deployment.

A grapevine yield prediction case study ties these elements together, showing how R’s modern tools can deliver reproducible, interpretable, and production-ready machine learning pipelines.