The goal of tune is to facilitate the tuning of hyper-parameters the tidymodels packages. It replies heavily on recipes, parsnip, and dials.
You can install the current development version using
devtools::install_github("tidymodels/tune")There are a few package vignettes, the first being the Getting Started document. Other examples resources:
- basic grid search
- Bayesian optimization example
- an advanced text mining example
- notes on optimizations and parallel processing
- details on acquisition function for scoring parameter combinations
This is the development version. While the APIs are pretty stable they are subject to change due to user feedback and other issues. Please give as much feedback as possible regarding the good and bad parts.
There are references to workflow objects in tune. We have another package being developed called workflows that will streamline some of the tidymodels APIs. The functions in this package related to workflows don't do much right now and are placeholders for future functions.
In this version, you can jointly tune pre-processing parameters (via a recipe) and model parameters. Once workflows is released, you can also optimize post-processing parameters at the same time (e.g. the probability threshold for two-class problems).