Total Tidy Tuning Techniques
summary
Ames housing data and tidymodels recipes feed into parsnip model specs where placeholders like tune() mark parameters to optimize (example: glmnet’s penalty and mixture, splines’ degrees_of_freedom). Grid search via tune_grid uses an rsample resampling object (default 10‑fold CV), returns per‑fold metrics and notes, and offers select_best/collect_metrics/finalize to splice winners back into workflow. Iterative Bayesian search (tune_bayes), error/warning capture, saving out‑of‑sample predictions, and parallel execution are supported.
AI-written summary · about
chapters (11)
- 0:00Intro
- 1:34The Ames Housing Data
- 2:05Pre-processing with {recipes}
- 5:38Creating Models with parsnip
- 8:55Tagging Parameters using {tune}
- 11:58Ingredients for Grid Search in tidymodels
- 15:52Plot Method Shows the Marginal Relationships
- 17:07Updating the original objects We can splice in the best settings to the recipes ar model using the finalize_family of functions
- 18:03Miscellaneous Notes
- 20:01Iterative Search via Bayesian Optimization
- 20:50Plot Method Can Show Iterations
related talks
22mModelling, Tidyverse Way
57mGood Practices for Applied Machine Learning
25mHyperparameter Tuning and Visualization in Deep Learning
18mtidymodels: Adventures in Rewriting a Modeling Pipeline
15mThe tidysynthesis R package
17mSelfTune: Tuning Cluster Managers
26mModel Selection with Python: An Introduction to Hyper Parameter Tuning
22mMulti-level Machine Learning-Guided Autotuning for Efficient Code Generation on a Deep(…)