cluster_similarity: Computes the similarity between two clusterings of the same...

In ramhiser/clusteval: Evaluation of Clustering Algorithms

Description

For two clusterings of the same data set, this function calculates the similarity statistic specified of the clusterings from the comemberships of the observations. Basically, the comembership is defined as the pairs of observations that are clustered together.

Usage

cluster_similarity(labels1, labels2, similarity = "adjusted_rand")

Arguments

labels1	a vector of n clustering labels
labels2	a vector of n clustering labels
similarity	the similarity statistic to calculate. See similarity_methods for a listing of available similarity methods. By default, the adjusted Rand index is used.

Details

To calculate the similarity, we compute the 2x2 contingency table, consisting of the following four cells:

n_11:

the number of observation pairs where both observations are comembers in both clusterings

n_10:

the number of observation pairs where the observations are comembers in the first clustering but not the second

n_01:

the number of observation pairs where the observations are comembers in the second clustering but not the first

n_00:

the number of observation pairs where neither pair are comembers in either clustering

Currently, we have implemented the following similarity statistics:

• Adjusted Rand index

• Dice coefficient

• Fowlkes-Mallows coefficient

• Jaccard coefficient

• Phi coefficient

• Rand index

• Rogers-Tanimoto coefficient

• Russel-Rao coefficient

• Sokal-Sneath coefficient

To compute the contingency table, we use the comembership_table function.

Value

the similarity between the two clusterings

Examples

# Notice that the number of comemberships is 'n choose 2'.
iris_kmeans <- kmeans(iris[, -5], centers = 3)$cluster
iris_hclust <- cutree(hclust(dist(iris[, -5])), k = 3)
cluster_similarity(iris_kmeans, iris_hclust)

ramhiser/clusteval documentation built on May 26, 2019, 10:07 p.m.