hammer

The goal of hammer is to provide an experimental variant of the mc_cv() function from rsample which takes into consideration case weights when sampling observations for the analysis set.

Installation

You can install the development version of hammer like so:

pak::pkg_install(mikemahoney218/hammer)

Example

The only function in the package is h_mc_cv(), a minor fork of the mc_cv() function from rsample. This function works more or less the same way as mc_cv(), and with default arguments returns identical objects (outside of an additional class added to objects from h_mc_cv()):

library(hammer)

hammer_output <- withr::with_seed(1107, h_mc_cv(Orange))
rsample_output <- withr::with_seed(1107, rsample::mc_cv(Orange))

class(hammer_output) <- class(rsample_output)

all.equal(hammer_output, rsample_output)
#> [1] TRUE

The only difference is a new argument, weights, which lets you specify a column in data that contains the probability of any given observation being selected for an analysis set. For instance, if we give the 10th observation a weight of 1 and every other observation a weight of 0, we can make it so that each of our analysis sets only contain the 10th observation:

example_data <- tibble::tibble(x = 1:10, w = c(rep(0, 9), 1))

weighted_hammer <- h_mc_cv(example_data, weights = w, prop = 0.1)

vapply(weighted_hammer$splits, function(x) rsample::analysis(x)$x, numeric(1))
#>  [1] 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10

This also works with stratification; for instance, we can pull the same trick by assigning the 9th and 10th observations positive weights and different strata, ensuring that our analysis sets will always contain only the 9th and 10th observations:

example_data <- tibble::tibble(
  x = 1:10, 
  w = c(rep(0, 8), 1, 1), 
  s = rep(1:2, 5)
)

weighted_hammer <- h_mc_cv(example_data, weights = w, strata = s, prop = 0.2)

vapply(weighted_hammer$splits, function(x) rsample::analysis(x)$x, numeric(2))
#>      [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10] [,11] [,12] [,13] [,14]
#> [1,]    9    9    9    9    9    9    9    9    9     9     9     9     9     9
#> [2,]   10   10   10   10   10   10   10   10   10    10    10    10    10    10
#>      [,15] [,16] [,17] [,18] [,19] [,20] [,21] [,22] [,23] [,24] [,25]
#> [1,]     9     9     9     9     9     9     9     9     9     9     9
#> [2,]    10    10    10    10    10    10    10    10    10    10    10