Skip to contents

Non-hierarchical clustering: Affinity Propagation

Source: R/nhclu_affprop.R
nhclu_affprop.Rd

This function performs non-hierarchical clustering using the Affinity Propagation algorithm.

Usage

nhclu_affprop(
  similarity,
  index = names(similarity)[3],
  seed = NULL,
  p = NA,
  q = NA,
  maxits = 1000,
  convits = 100,
  lam = 0.9,
  details = FALSE,
  nonoise = FALSE,
  K = NULL,
  prc = NULL,
  bimaxit = NULL,
  exact = NULL,
  algorithm_in_output = TRUE
)

Arguments

similarity

The output object from similarity() or dissimilarity_to_similarity(), or a dist object. If a data.frame is used, the first two columns should represent pairs of sites (or any pair of nodes), and the subsequent column(s) should contain the similarity indices.

index

The name or number of the similarity column to use. By default, the third column name of similarity is used.

seed

The seed for the random number generator used when nonoise = FALSE.

p

Input preference, which can be a vector specifying individual preferences for each data point. If scalar, the same value is used for all data points. If NA, exemplar preferences are initialized based on the distribution of non-Inf values in the similarity matrix, controlled by q.

q

If p = NA, exemplar preferences are initialized according to the distribution of non-Inf values in the similarity matrix. By default, the median is used. A value between 0 and 1 specifies the sample quantile, where q = 0.5 results in the median.

maxits

The maximum number of iterations to execute.

convits

The algorithm terminates if the exemplars do not change for convits iterations.

lam

The damping factor, a value in the range [0.5, 1). Higher values correspond to heavier damping, which may help prevent oscillations.

details

If TRUE, detailed information about the algorithm's progress is stored in the output object.

nonoise

If TRUE, disables the addition of a small amount of noise to the similarity object, which prevents degenerate cases.

K

The desired number of clusters. If not NULL, the function apclusterK is called.

prc

A parameter needed when K is not NULL. The algorithm stops if the number of clusters deviates by less than prc percent from the desired value K. Set to 0 to enforce exactly K clusters.

bimaxit

A parameter needed when K is not NULL. Specifies the maximum number of bisection steps to perform. No warning is issued if the number of clusters remains outside the desired range.

exact

A flag indicating whether to compute the initial preference range exactly.

algorithm_in_output

A boolean indicating whether to include the original output of apcluster in the result. Defaults to TRUE.

Value

A list of class bioregion.clusters with five slots:

  1. name: A character string containing the name of the algorithm.

  2. args: A list of input arguments as provided by the user.

  3. inputs: A list describing the characteristics of the clustering process.

  4. algorithm: A list of objects associated with the clustering procedure, such as original cluster objects (if algorithm_in_output = TRUE).

  5. clusters: A data.frame containing the clustering results.

If algorithm_in_output = TRUE, the algorithm slot includes the output of apcluster.

Details

This function is based on the apcluster package (apcluster).

References

Frey B & Dueck D (2007) Clustering by Passing Messages Between Data Points. Science 315, 972-976.

See also

For more details illustrated with a practical example, see the vignette: https://biorgeo.github.io/bioregion/articles/a4_2_non_hierarchical_clustering.html.

Associated functions: nhclu_clara nhclu_clarans nhclu_dbscan nhclu_kmeans nhclu_affprop

Author

Pierre Denelle (pierre.denelle@gmail.com)
Boris Leroy (leroy.boris@gmail.com)
Maxime Lenormand (maxime.lenormand@inrae.fr)

Examples

comat_1 <- matrix(sample(0:1000, size = 10*12, replace = TRUE,
prob = 1/1:1001), 10, 12)
rownames(comat_1) <- paste0("Site", 1:10)
colnames(comat_1) <- paste0("Species", 1:12)
comat_1 <- cbind(comat_1,
                 matrix(0, 10, 8,
                        dimnames = list(paste0("Site", 1:10),
                                        paste0("Species", 13:20))))
                                        
comat_2 <- matrix(sample(0:1000, 
                         size = 10*12, 
                         replace = TRUE, 
                         prob = 1/1:1001), 
                  10, 12)
rownames(comat_2) <- paste0("Site", 11:20)
colnames(comat_2) <- paste0("Species", 9:20)
comat_2 <- cbind(matrix(0, 10, 8, 
                        dimnames = list(paste0("Site", 11:20),
                                        paste0("Species", 1:8))),
                 comat_2)
                 
comat <- rbind(comat_1, comat_2)

dissim <- dissimilarity(comat, metric = "Simpson")
sim <- dissimilarity_to_similarity(dissim)

clust1 <- nhclu_affprop(sim)

clust2 <- nhclu_affprop(sim, q = 1)

# Fixed number of clusters 
clust3 <- nhclu_affprop(sim, K = 2, prc = 10, bimaxit = 20, exact = FALSE)
#> Trying p = 0.9930872 
#>    Number of clusters: 6 
#> Trying p = 0.9308716 
#>    Number of clusters: 6 
#> Trying p = 0.3087157 
#>    Number of clusters: 1 
#> Trying p = 0.6543579 (bisection step no. 1 )
#>    Number of clusters: 1 
#> Trying p = 0.8271789 (bisection step no. 2 )
#>    Number of clusters: 3 
#> Trying p = 0.7407684 (bisection step no. 3 )
#>    Number of clusters: 1 
#> Trying p = 0.7839737 (bisection step no. 4 )
#>    Number of clusters: 1 
#> Trying p = 0.8055763 (bisection step no. 5 )
#>    Number of clusters: 3 
#> Trying p = 0.794775 (bisection step no. 6 )
#>    Number of clusters: 2 
#> 
#> Number of clusters: 2 for p = 0.794775