In cont-limno/speCluster: Spectral Clustering in R

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "images/"
)

speCluster

This package wraps the code at:

https://github.com/cont-limno/SpectralClustering4Regions

Installation

You can install speCluster from github with:

# install.packages("devtools")
devtools::install_github("cont-limno/speCluster")

Usage

Load package

library(speCluster)

Read data

dt     <- read.csv("data-raw/dataTerrFW.csv", header = T)
coords <- read.csv("data-raw/latLong18876.csv", header = T)

Prep Input

Here we create a data matrix dt where each row is an observation and each column is a variable. In this example, we subset a larger matrix to focus on the US State of Missouri. Uninformative (constant) variables are removed with rm_constant_columns. Next, we create a constraint matrix that incorporates information on the spatial dependence among observations.

# Create data matrix
in_state <- as.character(dt$hu12_states) == "MO"
dt <- dt[in_state, -c(1,2)]
dt <- as.matrix(rm_constant_columns(dt))
colnames(dt) <- NULL

# Create constraint matrix
coords <- as.matrix(coords[in_state,])
colnames(coords) <- NULL
nb <- spdep::dnearneigh(coords, 0, 0.192)
nb <- nb_collapse(nb)
cmat <- neighborMatrix(nb, conFactor = 1)

Generate Clusters

results <- speCluster(data = dt, conMatrix = cmat, 
                      cluster.number = 10)
summary(results)
results$SS
head(results$clusters)
mapping(lat = coords[,1], long = coords[,2],
         clusters = results$clusters)

Generate Hierarchical Clusters

results <- hspeCluster(data = dt, conMatrix = cmat, 
                      cluster.number = 10)
summary(results)
results$SS
head(results$clusters)
mapping(lat = coords[,1], long = coords[,2],
         clusters = results$clusters[,ncol(results$clusters)])