blb: Bag of little bootstraps

In DrylandEcology/rSW2utils: Utility Tools for SOILWAT2 and STEPWAT2 Simulation Experiments

Bag of little bootstraps

Description

Bag of little bootstrap as described by Kleiner et al. 2014 implemented with adaptive convergence checking.

Usage

blb(
  data,
  subset_size_b = nrow(data)^0.7,
  n_subsets = NA,
  n_resamples = 100,
  window_subsets = 3,
  window_resamples = 20,
  epsilon = 0.05,
  fun_estimator = NULL,
  fun_metric = NULL
)

Arguments

data	A two-dimensional numerical data object.
subset_size_b	An integer value. The number of rows for each subset bootstraps. Kleiner et al. 2014 suggest empirically a value of nrow(data) ^ lambda with lambda = 0.7, but lambda between 0.1 and 1.
n_subsets	An integer value. The upper limit of sampled subsets s. If NA, then all subsets are sampled. If convergence is achieved earlier, then not all s subsets are processed.
n_resamples	An integer value. The upper limit of Monte-Carlo iterations (resamples, r) carried out on each subset. Kleiner et al. 2014 found empirically that a value of r = 100 worked well for confidence intervals. If convergence is achieved earlier, then not all r resamples are processed.
window_subsets	An integer value. The window size of the number of previous subsets to consider for adaptive convergence checking.
window_resamples	An integer value. The window size of the number of previous resamples to consider for adaptive convergence checking.
epsilon	A positive numerical value. The acceptable relative error to determine convergence.
fun_estimator	A function with two arguments x and weights where x will be j-th subset of data and weights a vector of counts defining the k-th resample of the j-th subset (see examples). The return value is a named vector of the estimator(s) of interest.
fun_metric	A function with one argument x which will be applied to each element of the results of fun_estimator across all resamples of the j-th subset. The return value is a named vector of the estimator(s) of quality assessment for the estimator(s) of interest.

Value

A two-dimensional object of means across all subsets where rows represent the estimator(s) of quality assessment, i.e., the output of fun_metric, and the columns represent the the estimator(s) of interest, i.e., the output of fun_estimator.

References

Kleiner, A., A. Talwalkar, P. Sarkar, and M. I. Jordan. 2014. A scalable bootstrap for massive data. Journal of the Royal Statistical Society: Series B (Statistical Methodology) 76:795–816.

See Also

Function drBLB of package https://github.com/delta-rho/datadr.

Examples

n <- 10000
xt <- seq(0, 10, length.out = n)
ex_data <- data.frame(
  x1 = sample(xt),
  x2 = sample(xt)
)

# Linear regression with coefficients 1, 2, and 3
ex_data[, "y"] <-
  1 + rnorm(n, 0, 1) + 2 * ex_data[, "x1"] + 3 * ex_data[, "x2"]

# Estimate coefficients with BLB
blb(
  data = ex_data,
  fun_estimator = function(x, weights) {
    coef(lm(
      y ~ x1 + x2,
      data = x,
      weights = weights / max(weights)
    ))
  },
  fun_metric = function(x) {
    quantile(x, probs = c(0.025, 0.5, 0.975))
  }
)

DrylandEcology/rSW2utils documentation built on Dec. 9, 2023, 10:44 p.m.