clust_describe: Describe clusters of an umapscan object

In juba/umapscan: UMAP and DBSCAN Pragmatic and Supervised clustering

Description

Describe clusters of an umapscan object

Usage

clust_describe(
  us,
  parent = "",
  labels = TRUE,
  type = c("boxplot", "ridges", "barplot", "keyness"),
  position = "fill",
  keyness_measure = c("chi2", "lr", "exact", "pmi"),
  n_terms = 20,
  text_size = 10,
  free_scale = TRUE,
  show_negative = TRUE
)

Arguments

us	umapscan object to describe clusters
parent	name of the parent cluster
labels	if TRUE, use cluster label instead of id if available
type	plot type, either "boxplot", "ridges", "barplot" or "keyness"
position	if type = "barplot", position argument to add to geom_bar
keyness_measure	if type = "keyness", passed as measure argument to quanteda::textstat_keyness
n_terms	if type = "keyness", number of terms to show
text_size	if type = "keyness", plot text size
free_scale	if type = "keyness", don't use the same scale on each barplot
show_negative	if TRUE and type = "keyness", show negative keyness features

Details

"boxplot" and "ridges" are suitable for continuous variables, "barplot" is better for binary categorical ones. "keyness" is suitable when the original data is a document-feature matrix.

See Also

clust_compute(), clust_get_data(), clust_membership(), clust_rename()

Examples

library(dplyr)
iris_num <- iris %>% select_if(is.numeric)
us <- new_umapscan(iris_num, n_neighbors = 25, min_dist = 0.1, seed = 1337)
us <- clust_compute(us, minPts = 3, eps = 0.5)
clust_describe(us)
us <- clust_compute(us, minPts = 3, eps = 0.45, parent = "3")
clust_describe(us, type = "ridge")
clust_describe(us, parent = "3")

juba/umapscan documentation built on Aug. 29, 2020, 7:31 a.m.