Skip to contents

Fit meta learner

Source: R/meta_learner.R
fit_meta_learner.Rd

This function subsets the full data by column subsamples (rate=50%) The optimal hyperparameter search is performed based on spatiotemporal cross-validation schemes. As of version 0.4.5, users can define metric used for selecting best hyperparameter set (default = "rmse").

Usage

fit_meta_learner(
  data,
  c_subsample = 0.5,
  r_subsample = 1,
  yvar = "Arithmetic.Mean",
  target_cols = c("site_id", "time", "lon", "lat", "Event.Type"),
  args_generate_cv = list(),
  tune_iter = 50L,
  trim_resamples = FALSE,
  return_best = TRUE,
  metric = "rmse"
)

Arguments

data

data.frame. Full data.frame of base learner predictions and AQS spatiotemporal identifiers. attach_pred.

c_subsample

numeric(1). Rate of column resampling. Default is 0.5.

r_subsample

numeric(1). The proportion of rows to be used. Default is 1.0, which uses full dataset but setting is required to balance groups generated with make_subdata

yvar

character(1). Outcome variable name

target_cols

characters(1). Columns from data to be retained during column resampling. Default is c("site_id", "time", "Event.Type", "lon", "lat").

args_generate_cv

List of arguments to be passed to switch_generate_cv_rset function.

tune_iter

integer(1). Bayesian optimization iterations. Default is 50.

trim_resamples

logical(1). Default is TRUE, which replaces the actual data.frames in splits column of tune_results object with NA. Passed to fit_base_tune.

return_best

logical(1). If TRUE, the best tuned model is returned. Passed to fit_base_tune.

metric

character(1). The metric to be used for selecting the best. Must be one of "rmse", "rsq", "mae". Default = "rmse". Passed to fit_base_tune.

Value

List of 3, including the best-fit model, the best hyperparameters, and the all performance records from tune::tune_bayes(). Note that the meta learner function returns the best-fit model, not predicted values.

See also

fit_base_tune make_subdata