summaryheat: Within-cluster homogeneity and between-cluster heterogeneity...

In lukadw11/Clusty: Evaluate clustering using distance heat maps

Description

This function takes a data frame of instances characterized by cluster IDs, condenses the instance vectors, creates a distance matrix between each instance grouped by cluster, and visualizes the clustering summary using a heat map gradient from very similar (blue) to very different (red). The diagonal is then ranked by most to least homogeneous cluster. The upper and lower symmetrical triangular matrices are ranked by most to least heterogeneous. The intuition is that a highly homogeneous cluster (i.e. with similar instances) corresponds to strong clustering, and likewise with highly differentiated cluster comparisons.

Usage

summaryheat(df, compare_metric = "Mean", ranks = TRUE, dist_metric = "euclidean", axislabs = "Aggregated Cluster Distance", title = "Differentiation Rank", interactive = FALSE)

Arguments

df	(REQUIRED) Data frame containing numeric features and a cluster ID column. Cluster ID column must be labelled "Cluster" and contain exclusive cluster IDs of type numeric, integer, or factor.
compare_metric	(OPTIONAL) Character argument for how the distance between cluster instances should be compared. The options are "mean","sd", "range", and "median". The assumption with each metric is that a higher value is good for differentiation (i.e. between-cluster heterogeneity) and a lower value is good for validating cluster membership (i.e. within-cluster homogeneity).
ranks	(OPTINAL) Boolean argument of whether or not to include ranks in the heat map output. Diagonal is ranked by within-cluster instance similarity. Upper and lower triangle are ranked by between-cluster instance differentiation.
dist_metric	(OPTIONAL) Character argument of what method to use for measuring distance between instances. Arguments are limited to those provided in the dist base-function, which include "euclidean", "maximum", "Manhattan", "Canberra", "binary" and "Malinowski".
axislabs	(OPTIONAL) Character argument of the plot axis labels (note: one argument)
title	(OPTIONAL) Character argument of the plot title.

Value

A ggplot cluster-level distance matrix heat map visualization.

Author(s)

Derek Lukacsko & Jonathan Bourne

See Also

bigextract, bigheat

Examples

# Data frame of features
df <- iris[,c(1:3)]

# Create clusters
k = 3
fit <- kmeans(df, k)

# Append cluster memebership to instance vectors
df$Cluster = fit$cluster

# Visualize using summaryheat
sumnmaryheat(df, compare_metric = "Median")

lukadw11/Clusty documentation built on May 21, 2019, 8:57 a.m.