Good Practices for Applied Machine Learning
summary
Tidymodels is a unified R framework that packages recipes (step_date, step_normalize, genre parsing), workflows (bind a recipe to a model like Cubist), and tuneable hyperparameter tools so preprocessing is saved and reapplied; deployment uses vetiver to version models and a one‑line Plumber call to spin up an API. The racing tuner drops candidates to ~7% of 50 configs, yielding ~3x speed with parallelism (~9x without). Supervised feature selection must run inside resampling (10 folds, final fit on all data).
AI-written summary · about
chapters (10)
- 0:00Introduction to tidy models
- 1:34Philosophy and design
- 5:46Real-world modeling workflow
- 10:37Ergonomics and recipes
- 18:33Deployment with vetiver
- 26:07Efficiency and effectiveness
- 31:56Extensibility and modularity
- 36:04Safety in machine learning
- 43:28Conclusion and resources
- 50:05Q&A session
related talks
57mGood Practices for Applied Machine Learning
22mModelling, Tidyverse Way
18mtidymodels: Adventures in Rewriting a Modeling Pipeline
21mTidypredict with recipes, turn workflow to SQL, spark, duckdb and beyond
26mModel Selection with Python: An Introduction to Hyper Parameter Tuning
50mBuilding an MLOps strategy from the ground up
36mApplied Machine Learning Playbook: From Idea to Production
20mBetter Care and Feeding of Machine Learning Models