Tune hyperparameters using random search

Usage

tl_tune_random(
  data,
  formula,
  method,
  param_space,
  n_iter = 10,
  folds = 5,
  metric = NULL,
  maximize = NULL,
  verbose = TRUE,
  seed = NULL,
  ...
)

Arguments

data

A data frame containing the training data

formula

A formula specifying the model

method

The modeling method to tune

param_space

A named list of parameter spaces to sample from. Each element is read by its type and length:

a function: called with no arguments to draw one value
a list: a set of candidates, each drawn whole, e.g. list(c(10), c(20, 10)) for hidden_layers
c(min, max, "log"): log-uniform draw between min and max
a single value: used as given in every iteration
two whole numbers: integer range, e.g. c(10, 20) draws from 10:20
three or more numbers: a discrete set, sampled from as given, whether or not they are whole
two other numbers: uniform draw between them, e.g. c(0.01, 0.1)
character or factor: categorical, sampled from as given
logical: sampled from the values given

n_iter

Number of random parameter combinations to try

folds

Number of cross-validation folds

metric

Metric to optimize

maximize

Logical; whether to maximize (TRUE) or minimize (FALSE) the metric

verbose

Logical; whether to print progress

seed

Random seed for reproducibility

...

Additional arguments passed to tl_model

Value

A tidylearn model object fitted with the best hyperparameters. Tuning results are stored as an attribute "tuning_results", a list containing param_space, results, best_params, best_metric, metric, and maximize.

results has one row per iteration: iteration, mean_metric, n_folds_ok, and a column per parameter, as described for tl_tune_grid. The best parameters are chosen by the same rules, and mtry is capped the same way; duplicate draws are kept, so there are always n_iter rows.

Examples

# \donttest{
model <- tl_tune_random(mtcars, mpg ~ ., method = "tree",
  param_space = list(cp = c(0.01, 0.1), minsplit = c(10, 20)),
  n_iter = 3, folds = 2, verbose = FALSE)
# }