find_knn {knn.covertree}

kNN search

Description

k nearest neighbor search with custom distance function.

Usage

find_knn(data, k, ..., query = NULL, distance = c("euclidean",
  "cosine", "rankcor"), sym = TRUE)

Arguments

data

Data matrix

k

Number of nearest neighbors

...

Unused. All parameters to the right of the ... have to be specified by name (e.g. find_knn(data, k, distance = 'cosine'))

query

Query matrix. In knn and knn_asym, query and data are identical

distance

Distance metric to use. Allowed measures: Euclidean distance (default), cosine distance (1-corr(c_1, c_2)) or rank correlation distance (1-corr(rank(c_1), rank(c_2)))

sym

Return a symmetric matrix (as long as query is NULL)?

Value

A list with the entries:

index

A nrow(data) \times k integer matrix containing the indices of the k nearest neighbors for each cell.

dist

A nrow(data) \times k double matrix containing the distances to the k nearest neighbors for each cell.

dist_mat

A dgCMatrix if sym == TRUE, else a dsCMatrix (nrow(query) \times nrow(data)). Any zero in the matrix (except for the diagonal) indicates that the cells in the corresponding pair are close neighbors.

Examples

# The default: symmetricised pairwise distances between all rows
pairwise <- find_knn(mtcars, 5L)
image(as.matrix(pairwise$dist_mat))

# Nearest neighbors of a subset within all
mercedeses <- grepl('Merc', rownames(mtcars))
merc_vs_all <- find_knn(mtcars, 5L, query = mtcars[mercedeses, ])
# Replace row index matrix with row name matrix
matrix(
  rownames(mtcars)[merc_vs_all$index],
  nrow(merc_vs_all$index),
  dimnames = list(rownames(merc_vs_all$index), NULL)
)[, -1]  # 1st nearest neighbor is always the same row