cluster_data: Clustering via Dissimilarity Matrix based on String Distances

View source: R/clusters.R

cluster_dataR Documentation

Clustering via Dissimilarity Matrix based on String Distances

Description

Performs clustering using specified dissimilarity measures and clustering methods. The rows of the data are first converted to strings and compared using the dissimilarity measures available in the stringdist package.

Usage

cluster_data(
  data,
  k,
  dissimilarity = "hamming",
  method = "pam",
  na_syms = c("*", "%"),
  weighted = FALSE,
  lambda = 1,
  ...
)

cluster_sequences(
  data,
  k,
  dissimilarity = "hamming",
  method = "pam",
  na_syms = c("*", "%"),
  weighted = FALSE,
  lambda = 1,
  ...
)

Arguments

data

A data.frame, a matrix in wide format, or a tna_data object (in which case, the sequence data is extracted automatically).

k

An integer giving the number of clusters.

dissimilarity

A character string specifying the dissimilarity measure. The available options are: "osa", "lv", "dl", "hamming", "lcs", "qgram", "cosine", "jaccard", and "jw". See stringdist::stringdist-metrics for more information on these measures.

method

A character string specifying clustering method. The available methods are "pam", "ward.D", "ward.D2", "complete","average", "single", "mcquitty", "median", and "centroid". See cluster::pam() and stats::hclust() for more information on these methods.

na_syms

A character vector of symbols or factor levels to convert to explicit missing values.

weighted

A logical value indicating whether the dissimilarity measure should be weighted (the default is FALSE for no weighting). If TRUE, earlier observations of the sequences receive a greater weight in the distance calculation with an exponential decay. Currently only supported for the Hamming distance.

lambda

A numeric value defining the strength of the decay when weighted = TRUE. The default is 1.0.

...

Additional arguments passed to stringdist::stringdist().

Value

A tna_clustering object which is a list containing:

  • data: The original data.

  • k: The number of clusters.

  • assignments: An integer vector of cluster assignments, one per sequence (row of data).

  • silhouette: The average silhouette width of the clustering.

  • sizes: An integer vector of cluster sizes.

  • method: The clustering method used.

  • distance: The dissimilarity matrix (a dist object).

The result can be passed to group_model() to build a TNA model for each cluster.

References

Kaufman, L., & Rousseeuw, P. J. (1990). Finding Groups in Data: An Introduction to Cluster Analysis. Wiley. \Sexpr[results=rd]{tools:::Rd_expr_doi("10.1002/9780470316801")}

Rousseeuw, P. J. (1987). Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics, 20, 53–65. \Sexpr[results=rd]{tools:::Rd_expr_doi("10.1016/0377-0427(87)90125-7")}

See Also

Cluster-related functions communities(), group_model(), mmm_stats(), print.tna_clustering(), rename_groups()

Examples

data <- data.frame(
  T1 = c("A", "B", "A", "C", "A", "B"),
  T2 = c("B", "A", "B", "A", "C", "A"),
  T3 = c("C", "C", "A", "B", "B", "C")
)

# PAM clustering with the Hamming distance (the defaults)
result <- cluster_sequences(data, k = 2)
print(result)

# Hierarchical clustering with the optimal string alignment distance
result <- cluster_sequences(data, k = 2, dissimilarity = "osa",
                            method = "complete")


tna documentation built on Nov. 17, 2026, 1:06 a.m.