Control aspects of the Bayesian search process
control_bayes(
verbose = FALSE,
no_improve = 10L,
uncertain = Inf,
seed = sample.int(10^5, 1),
extract = NULL,
save_pred = FALSE,
time_limit = NA,
pkgs = NULL,
save_workflow = FALSE,
save_gp_scoring = FALSE,
event_level = "first",
parallel_over = NULL
)A logical for logging results (other than warnings and errors,
which are always shown) as they are generated during training in a single
R process. When using most parallel backends, this argument typically will
not result in any logging. If using a dark IDE theme, some logging messages
might be hard to see; try setting the tidymodels.dark option with
options(tidymodels.dark = TRUE) to print lighter colors.
The integer cutoff for the number of iterations without better results.
The number of iterations with no improvement before an
uncertainty sample is created where a sample with high predicted variance is
chosen (i.e., in a region that has not yet been explored). The iteration
counter is reset after each uncertainty sample. For example, if uncertain = 10, this condition is triggered every 10 samples with no improvement.
An integer for controlling the random number stream.
An optional function with at least one argument (or NULL)
that can be used to retain arbitrary objects from the model fit object,
recipe, or other elements of the workflow.
A logical for whether the out-of-sample predictions should be saved for each model evaluated.
A number for the minimum number of minutes (elapsed) that
the function should execute. The elapsed time is evaluated at internal
checkpoints and, if over time, the results at that time are returned (with
a warning). This means that the time_limit is not an exact limit, but a
minimum time limit.
An optional character string of R package names that should be loaded (by namespace) during parallel processing.
A logical for whether the workflow should be appended to the output as an attribute.
A logical to save the intermediate Gaussian process
models for each iteration of the search. These are saved to
tempdir() with names gp_candidates_{i}.RData where i is the iteration.
These results are deleted when the R session ends. This option is only
useful for teaching purposes.
A single string containing either "first" or "second".
This argument is passed on to yardstick metric functions when any type
of class prediction is made, and specifies which level of the outcome
is considered the "event".
A single string containing either "resamples" or
"everything" describing how to use parallel processing. Alternatively,
NULL is allowed, which chooses between "resamples" and "everything"
automatically.
If "resamples", then tuning will be performed in parallel over resamples
alone. Within each resample, the preprocessor (i.e. recipe or formula) is
processed once, and is then reused across all models that need to be fit.
If "everything", then tuning will be performed in parallel at two levels.
An outer parallel loop will iterate over resamples. Additionally, an
inner parallel loop will iterate over all unique combinations of
preprocessor and model tuning parameters for that specific resample. This
will result in the preprocessor being re-processed multiple times, but
can be faster if that processing is extremely fast.
If NULL, chooses "resamples" if there are more than one resample,
otherwise chooses "everything" to attempt to maximize core utilization.
Note that switching between parallel_over strategies is not guaranteed
to use the same random number generation schemes. However, re-tuning a
model using the same parallel_over strategy is guaranteed to be
reproducible between runs.
For extract, this function can be used to output the model object, the
recipe (if used), or some components of either or both. When evaluated, the
function's sole argument has a fitted workflow If the formula method is used,
the recipe element will be NULL.
The results of the extract function are added to a list column in the
output called .extracts. Each element of this list is a tibble with tuning
parameter column and a list column (also called .extracts) that contains
the results of the function. If no extraction function is used, there is no
.extracts column in the resulting object. See tune_bayes() for more
specific details.
Note that for collect_predictions(), it is possible that each row of the
original data point might be represented multiple times per tuning
parameter. For example, if the bootstrap or repeated cross-validation are
used, there will be multiple rows since the sample data point has been
evaluated multiple times. This may cause issues when merging the predictions
with the original data.