vertex_covariate_dist: Computes covariate distance between connected vertices

Description

Computes covariate distance between connected vertices

Usage

vertex_covariate_dist(graph, X, p = 2)
vertex_mahalanobis_dist(graph, X, S)

Arguments

graph

A square matrix of size $n$ of class dgCMatrix.

A numeric matrix of size $n * K$. Vertices attributes

Numeric scalar. Norm to compute

Square matrix of size ncol(x). Usually the var-covar matrix.

Value

A matrix of size $n * n$ of class dgCMatrix. Will be symmetric only if graph is symmetric.

Details

Faster than dist, these functions compute distance metrics between pairs of vertices that are connected (otherwise skip).

The function vertex_covariate_dist is the simil of dist and returns p-norms. It is implemented as follows (for each pair of vertices):

$D_{i j} = {(\sum_{k = 1}^{K} (X_{i k} - X_{j k})^{p})}^{1 / p} if g r a p h_{i, j} \neq 0$

In the case of mahalanobis distance, for each pair of vertex $(i,j)$, the distance is computed as follows:

$D_{i j} = {((X_{i} - X_{j}) \times S \times (X_{i} - X_{j})^{'})}^{1 / 2} if g r a p h_{i, j} \neq 0$

References

Mahalanobis distance. (2016, September 27). In Wikipedia, The Free Encyclopedia. Retrieved 20:31, September 27, 2016, from https://en.wikipedia.org/w/index.php?title=Mahalanobis_distance&oldid=741488252

Examples

Run this code

# Distance (aka p norm) -----------------------------------------------------
set.seed(123)
G <- rgraph_ws(20, 4, .1)
X <- matrix(runif(40), ncol=2)

vertex_covariate_dist(G, X)

# Mahalanobis distance ------------------------------------------------------
S <- var(X)

M <- vertex_mahalanobis_dist(G, X, S)

# Example with diffnet objects ----------------------------------------------

data(medInnovationsDiffNet)
X <- cbind(
  medInnovationsDiffNet[["proage"]],
  medInnovationsDiffNet[["attend"]]
)

S <- var(X, na.rm=TRUE)
ans <- vertex_mahalanobis_dist(medInnovationsDiffNet, X, S)

Run the code above in your browser using DataLab

Get 50% off unlimited learning