earthmover

The earthmover distance calculates distances between sets of data. It is calculated as the sum of the Minkowski distances of samples, weighted by their transport coefficient. This package implements the classic earthmover distance, the earthmover distance for samples that are embedded by a model, including random forests, as well as analyses and visualizations to assess the stability of the distances with respect to their constituent samples.

Installation

You can install the development version of earthmover from GitHub with:

# install.packages("devtools")
devtools::install_github("kaneplusplus/earthmover")

Example

library(earthmover)
library(future)
library(tibble)

# Run in parallel.
plan("multisession", workers = 2)

# Subset iris for our x and y.
iris1 = iris[1:50,]
iris2 = iris[51:100,]

# Formatting output for nicer presentation.
pairs_to_tibble = \(x) {
  x$pairs = as_tibble(x$pairs, .name_repair = "minimal")
  x
}

# The distance between the two subsets of data.
emd(iris1, iris2) |> 
  pairs_to_tibble()
#> $dist
#> [1] 3.541949
#> 
#> $pairs
#> # A tibble: 50 × 3
#>       ``    ``    ``
#>    <dbl> <dbl> <dbl>
#>  1     1    14  0.02
#>  2     2    19  0.02
#>  3     3    31  0.02
#>  4     4    40  0.02
#>  5     5    47  0.02
#>  6     6    28  0.02
#>  7     7    35  0.02
#>  8     8    18  0.02
#>  9     9    44  0.02
#> 10    10    13  0.02
#> # ℹ 40 more rows

# The distance between the a subsets and itself.
emd(iris1, iris1) |>
  pairs_to_tibble()
#> $dist
#> [1] 0
#> 
#> $pairs
#> # A tibble: 50 × 3
#>       ``    ``    ``
#>    <dbl> <dbl> <dbl>
#>  1     1     1  0.02
#>  2     2     2  0.02
#>  3     3     3  0.02
#>  4     4     4  0.02
#>  5     5     5  0.02
#>  6     6     6  0.02
#>  7     7     7  0.02
#>  8     8     8  0.02
#>  9     9     9  0.02
#> 10    10    10  0.02
#> # ℹ 40 more rows