distance: Calculate dissimilarity/distance metrics

Description

This function calculates a variety of dissimilarity or distance metrics. Although it duplicates the functionality of dist() and bcdist(), it is written in such a way that new metrics can easily be added. distance() was written for extensibility and understandability, and is not necessarily an efficient choice for use with large matrices.

Usage

distance(x, method = "euclidean", sprange=NULL, spweight=NULL, icov, inverted = FALSE)

Value

Returns a lower-triangular distance matrix as an object of class "dist".

Arguments

x: matrix or data frame with rows as samples and columns as variables (such as species). Distances will be calculated for each pair of rows.
method: Currently 7 dissimilarity metrics can be calculated: "euclidean", "bray-curtis", "manhattan", "mahalanobis" (squared Mahalanobis distance), "jaccard", "difference", "sorensen", "gower", "modgower10" (modified Gower, base 10), "modgower2" (modified Gower, base 2). Partial matching will work for selecting a method.
sprange: Gower dissimilarities offer the option of dividing by the species range. If sprange=NULL no range is used. If sprange is a vector of length nrow(x) it is used for standardizing the dissimilarities.
spweight: Euclidean, Manhattan, and Gower dissimilarities allow weighting. If spweight=NULL, no weighting is used. If spweight="absence", then W=0 if both species are absent and 1 otherwise, thus deleting joint absences.
icov: Optional covariance matrix; only used if method="mahalanobis" since Mahalanobis distance requires calculating the variance-covariance matrix for the entire dataset. Providing icov directly makes it possible to calculate distances for a subset of the full dataset.
inverted: If TRUE, the optional covariance matrix for method="mahalanobis" is not inverted before solving. Providing an inverted matrix may speed up calculations.

Author

Sarah Goslee

Examples

Run this code

data(iris)
iris.bc <- distance(iris[, 1:4], "bray-curtis")

# The effect of specifying icov:

# calculate Mahalanobis distance for the full iris dataset
iris.md <- full(distance(iris[, 1:4], "mahal"))
iris.md[1, 2] # Mahalanobis distance between samples 1 and 2 

# calculate Mahalanobis for just one species
setosa.md <- full(distance(iris[iris$Species == "setosa", 1:4], "mahal"))
setosa.md[1, 2] # Mahalanobis distance between samples 1 and 2 

# use the covariance matrix for the full dataset to scale for one species
setosa.scaled.md <- full(distance(iris[iris$Species == "setosa", 1:4],
  "mahal", icov=var(iris[,1:4])))
setosa.scaled.md[1, 2] # Mahalanobis distance between samples 1 and 2