Skip to content

coda.base

Overview

coda.base provides a compact and efficient toolkit for compositional data analysis using log-ratio coordinates. It supports the complete workflow from compositions to coordinates and back, including:

  • transforming compositions with coordinates() and composition();
  • constructing ALR, CLR, ILR, principal-component and balance bases;
  • applying closure, perturbation and powering;
  • computing Aitchison and other compositional distances;
  • replacing rounded zeros and missing values; and
  • finding principal balances with exact, constrained or tabu-search methods.

Core numerical operations are implemented in C++ for performance.

Installation

Install the released version from CRAN:

install.packages("coda.base")

Install the development version from GitHub:

# install.packages("pak")
pak::pak("mcomas/coda.base")

Usage

Transform a composition into isometric log-ratio coordinates and reconstruct the original composition:

library(coda.base)

x <- c(a = 0.2, b = 0.3, c = 0.5)

B <- ilr_basis(x)
z <- coordinates(x, B)
z
#>       ilr1       ilr2
#> -0.2867071 -0.5826178

composition(z, B)
#>   a   b   c
#> 0.2 0.3 0.5

The package also provides the standard operations of the Aitchison geometry:

x <- c(a = 2, b = 3, c = 5)
y <- c(a = 1, b = 4, c = 5)

closure(x)
#>   a   b   c
#> 0.2 0.3 0.5
perturbation(closure(x), closure(y))
#> [1] 0.05128205 0.30769231 0.64102564
dist_coda(rbind(x, y))
#>           x
#> y 0.7130311

Principal balances provide interpretable coordinates that seek to capture the variation in a compositional data set:

votes <- parliament2017[-1]
B_pb <- pb_basis(votes, method = "constrained")
head(coordinates(votes, B_pb), 3)
#>         pb1       pb2       pb3       pb4       pb5       pb6       pb7
#> 1 -1.497066 0.8542554 -1.638509 0.1469780 -1.552474 -1.331674 0.5706805
#> 2 -1.347042 0.7209798 -1.531162 0.2267969 -1.790182 -1.415831 0.5581286
#> 3 -1.462949 1.0246594 -1.743697 0.1235819 -1.330296 -1.198655 0.4131829

Learn more

Getting help

If you find a bug, please open an issue with a minimal reproducible example.

Citation

If you use coda.base in research, retrieve the recommended citation with:

citation("coda.base")