postcard (version 1.1.0)

fit_best_learner: Find the best learner in terms of RMSE among specified learners using cross validation

Description

Find the best learner in terms of RMSE among specified learners using cross validation

Usage

fit_best_learner(
  preproc,
  data,
  cv_folds = 5,
  learners = default_learners(),
  verbose = options::opt("verbose")
)

Value

a trained workflow

Arguments

preproc

A list (preferably named) with preprocessing objects: formulas, recipes, or workflows::workflow_variables(). Passed to workflowsets::workflow_set().

data

A data frame.

cv_folds

a numeric with the number of cross-validation folds used when fitting and evaluating models

learners

a list (preferably named) containing named lists of elements model and optionally grid. The model element should be a parsnip model specification, which is passed to workflowsets::workflow_set as the model argument, while the grid element is passed as the grid argument of workflowsets::option_add

verbose

numeric verbosity level. Higher values means more information is printed in console. A value of 0 means nothing is printed to console during execution (Defaults to 2, overwritable using option 'postcard.verbose' or environment variable 'R_POSTCARD_VERBOSE')

Details

Ensure data compatibility with the learners.

See Also

See rctglm_with_prognosticscore() for a function that utilises this function to perform prognostic covariate adjustment.

Examples

# Generate some synthetic 2-armed RCT data along with historical controls
n <- 100
dat_rct <- glm_data(
  Y ~ 1+2*x1+3*a,
  x1 = rnorm(n, 2),
  a = rbinom (n, 1, .5),
  family = gaussian()
)
dat_hist <- glm_data(
  Y ~ 1+2*x1,
  x1 = rnorm(n, 2),
  family = gaussian()
)

# Fit a learner to the historical control data
learners <- list(
  mars = list(
    model = parsnip::set_engine(
      parsnip::mars(
        mode = "regression", prod_degree = 3
      ),
      "earth"
    )
  )
)
fit <- fit_best_learner(
  preproc = list(mod = Y ~ .),
  data = dat_hist,
  learners = learners
)

# Use it fx. to predict the "control outcome" in the 2-armed RCT
predict(fit, new_data = dat_rct)