Package 'cord'

Community estimation in G-models via CORD

Description

Partition data points (variables) into clusters/communities. Reference: Bunea et al (2020). Model assisted variable clustering: Minimax-optimal recovery and algorithms, Annals of Statistics, doi:10.1214/18-AOS1794.

Usage

cord(
  X,
  tau = 2 * sqrt(log(ncol(X))/nrow(X)),
  kendall = T,
  input = c("data", "cor", "dist")
)

Arguments

X	Input data matrix. It should be an n (samples) by p (variables) matrix when input is set to the value "data" by default. It can also be a p by p symmetric matrix when X is a correlation matrix or a distance matrix if input is set accordingly.
tau	Threshold to use at each iteration. A theoretical choice is about $2n^{-1/2}\log^{1/2} p$.
kendall	Whether to compute Kendall's tau correlation matrix from X, when input is set to "data". If FALSE, Pearson's correlation will be computed, usually faster for large p.
input	Type of input X. It should be set to "data" when X is an n (samples) by p (variables) matrix. If X is a correlation matrix or a distance matrix, it should be set to "cor" or "dist" respectively.

Value

list with one element: a vector of integers showing which cluster/community each point is assigned to.

Examples

set.seed(100)
X <- 2*matrix(rnorm(200*2), 200, 10)+matrix(rnorm(200*10), 200, 10)
cord(X)

---