tune: Hyperparameter tuning on rsample objects

In stevenpawley/tidycrossval: Hyperparameter Tuning Using Tidymodel Principles

Description

'tune“ is a generic function that accepts a tibble of resampling partitions generated by the resampling schemes available in the 'rsample' package. A 'parsnip' model specification and a 'recipes' recipe also need to be supplied to this function, along with a 'yardstick' scoring function. The 'param_grid' function accepts a 'grid_regular' or 'grid_random' object of tuning parameters generated by the 'dials' package. Hyperparameter tuning is then performed on this object, which is returned as a tibble with an additional list-column called 'tune_scores' containing the tuning scores.

Usage

tune(resamples, object, recipe = NULL, param_grid, scoring,
  maximize = TRUE, .options = future_options())

Arguments

resamples	tibble generated by rsample containing resamplign scheme
object	parsnip model specification, or a pipeline object
recipe	recipe object, optional if using a pipeline
param_grid	dials grid_regular or grid_random object
scoring	yardstick scoring function
maximize	logical, maximize the scoring function, default = TRUE
.options	future_options to pass additional packages required by tuning and fitting functions

Details

If the resampling object represents a 'nested_cv' object, then hyperparameter tuning is performed on the inner resampling partitions, and best best hyperparameters per outer fold are also returned as additional columns in the output tibble.

Value

tibble containing resampling results

stevenpawley/tidycrossval documentation built on Oct. 3, 2019, 3:32 p.m.