nn: Nearest neighbors and distances

In talegari/disto: Unified Interface to Distance, Dissimilarity, Similarity Matrices

Description

Obtain nearest neighbors and distances from a matrix or disto handle. k nearest or fixed radius neighbors are supported

Usage

nn(x, k, r, method = "euclidean", ...)

Arguments

x	Object of class 'disto' or a numeric matrix
k	Number of nearest neighbors
r	Radius for nearest neighbors
method	(string or function) distance metric when x is a matrix. Passed to 'proxy::dist'. Ignored when x is not a matrix.
...	Additional arguments for dapply when x is 'disto' object. Else additional arguments are sent to pbmclapply

Details

Exactly one among k or r has to be provided

Value

Object of class nn. A list with these elements:

• triplet: Matrix with three columns: row, col and distance. For a fixed observation(value in 'row'), all corresponding values in 'col' are the indexes of the nearest neighbors. All corresponding values in 'distance' are the distances to those nearest neighbors

• size: Size of the distance matrix or number of rows of the matrix

• k or r : Depending on the input

Examples

## Not run: 
# create a matrix
set.seed(100)
mat <- cbind(rnorm(3e3), rpois(3e3, 1))

# compute a distance matrix and get a disto handle
do <- stats::dist(mat)
dio <- disto(objectname = "do")

# nearest neighbors: k nearest and fixed radius
nn(dio, k = 1)
nn(mat, k = 1) # distance method defaults to 'euclidean'
str(nn(mat, k = 1)) # observe the structure of the output

nn(dio, r = 0.1)
nn(mat, r = 0.1)

# nearest neighbors parallelized: k nearest and fixed radius
# fast computation, higher memory usage
nn(dio, k = 1, nproc = 2)
nn(mat, k = 1, mc.cores = 2)

nn(dio, r = 0.1, nproc = 2)
nn(mat, r = 0.1, mc.cores = 2)

# different distance method
do <- stats::dist(mat, method = "manhattan")

nn(dio, k = 1, nproc = 2)
nn(mat, k = 1, method = "manhattan", mc.cores = 2)

nn(dio, r = 0.1, nproc = 2)
nn(mat, r = 0.1, method = "manhattan", mc.cores = 2)

## End(Not run)

talegari/disto documentation built on May 10, 2019, 3:19 a.m.