Skip to contents

Prepare indicators for bootstrap and cross validation

Source: R/prepare_indicators_bootstrap.R
prepare_indicators_bootstrap.Rd

Create the parameter lists required to perform bootstrapping (dubicube::bootstrap_cube()), calculate confidence intervals (dubicube::calculate_bootstrap_ci()), and perform cross-validation (dubicube::cross_validate_cube()).

This function is primarily intended for internal use by indicator functions, but can also be used directly to construct parameter lists for manual calls to the underlying dubicube functions.

Default bootstrap and confidence interval arguments are internally defined and can be modified via boot_args and ci_args. User-supplied values override defaults.

Usage

prepare_indicators_bootstrap(
  impact_cube_data,
  indicator = c("overall", "site", "species"),
  indicator_method,
  grouping_var = "year",
  ci_type = "perc",
  confidence_level = 0.95,
  boot_args = list(samples = 1000, seed = NA),
  ci_args = list(no_bias = TRUE),
  out_var = "taxonKey"
)

Arguments

impact_cube_data

An impact cube object (class "impact_cube") created with create_impact_cube_data().

indicator

Character string specifying the impact indicator to be computed. Options are "overall", "site", and "species".

indicator_method

Character string specifying the method used to compute the impact indicator (see methods in compute_impact_indicator()).

grouping_var

A character vector specifying the grouping variable(s) for the bootstrap and confidence interval calculations. The function supplied to bootstrap_cube() must return one row per group. The specified variables must not be redundant (e.g., "time_point" and "year" should not both be used if one is simply an alternative encoding of the other).

ci_type

Character string specifying the type of confidence interval to compute. Options include:

  • "perc": Percentile interval (default)

  • "bca": Bias-corrected and accelerated interval

  • "norm": Normal approximation interval

  • "basic": Basic bootstrap interval

confidence_level

Numeric value specifying the confidence level for the intervals. Default is 0.95 (95% confidence level). This value is passed internally as conf to calculate_bootstrap_ci().

boot_args

Named list of additional arguments passed to dubicube::bootstrap_cube(). Defaults are:

samples

Number of bootstrap replicates (default 1000).

seed

Seed for reproducibility (default NA, meaning no call to set.seed()).

User-supplied values override these defaults. Arguments that are internally defined in this function (e.g., data_cube, fun, indicator_method, grouping_var, processed_cube, method) must not be supplied via boot_args.

ci_args

Named list of additional arguments passed to dubicube::calculate_bootstrap_ci(). Default is:

no_bias

Logical; if TRUE, intervals are centered around the original estimates (bias is ignored).

User-supplied values override defaults. The arguments grouping_var, type, and conf are controlled via grouping_var, ci_type, and confidence_level, respectively, and must not be supplied via ci_args.

out_var

Character string specifying the column used for leave-one-out cross-validation. Default is "taxonKey", which enables leave-one-species-out cross-validation.

Value

A named list with three components:

bootstrap_params

List of parameters for dubicube::bootstrap_cube().

ci_params

List of parameters for dubicube::calculate_bootstrap_ci().

cv_params

List of parameters for dubicube::cross_validate_cube().

Examples

if (FALSE) { # \dontrun{
library(b3gbi)

acacia_cube <- process_cube(
  cube_name = cube_acacia_SA,
  grid_type = "eqdgc",
  first_year = 2010,
  last_year = 2024
)

impact_cube <- create_impact_cube_data(
  cube_data = acacia_cube,
  impact_data = eicat_acacia
)

params <- prepare_indicators_bootstrap(
  impact_cube_data = impact_cube,
  indicator = "overall",
  indicator_method = "mean_cum",
  boot_args = list(samples = 2000),
  ci_args = list(no_bias = FALSE)
)

# Bootstrap
bootstrap_results <- do.call(
  dubicube::bootstrap_cube,
  params$bootstrap_params
)

# Confidence intervals
ci_result <- do.call(
  dubicube::calculate_bootstrap_ci,
  c(bootstrap_results = list(bootstrap_results), params$ci_params)
)

# Cross-validation
cv_results <- do.call(
  dubicube::cross_validate_cube,
  params$cv_params
)
} # }