Skip to contents

Transduce hierarchical clustering nodes to shapes

Source: R/stat_hclust.R
transduce.stat_hclust.Rd

Reconstruct shapes at cluster centers or at any node in the tree.

Usage

# S3 method for class 'stat_hclust'
transduce(object, positions)

Arguments

object

A stat_hclust object

positions

A tibble specifying positions. Can contain:

  • cluster column: cluster numbers (if tree was cut)

  • node column: node numbers in the tree (tips are 1:n, internal nodes are n+1:...)

Value

A tibble with reconstructed coefficients and shapes

Details

For hierarchical clustering, transduce can reconstruct shapes at:

Cluster centers (if tree was cut with k or h):

transduce(hc, tibble(cluster = 1:4))

Returns the mean shape of each cluster.

Any node in the tree:

transduce(hc, tibble(node = c(45, 50, 55)))

Returns the mean shape of all descendants of that node. This allows visualization of "ancestral" shapes at internal nodes.

Node numbering follows stats::hclust() convention:

  • Nodes 1 to n: original observations (tips)

  • Nodes n+1 to 2n-1: internal nodes from merges

Examples

if (FALSE) { # \dontrun{
hc <- boteft %>% stat_hclust(k = 4)

# Cluster centers
centers <- transduce(hc, tibble(cluster = 1:4))

# Shapes at internal nodes
hc2 <- boteft %>% stat_hclust()
internal <- transduce(hc2, tibble(node = c(45, 50, 55, 60)))

# Root node (all data)
root <- transduce(hc2, tibble(node = 2 * nrow(boteft) - 1))
} # }