add.distance {sperrorest}

Add distance information to resampling objects

Description

Add distance information to resampling objects

Usage

add.distance(object, ...)

## S3 method for class 'resampling'
add.distance(object, data, coords = c("x", "y"), ...)

## S3 method for class 'represampling'
add.distance(object, data, coords = c("x", "y"), mode = "future", ...)

Arguments

object

resampling or represampling object.

...

Additional arguments to dataset_distance and add.distance.resampling, respectively.

data

data.frame containing at least the columns specified by coords

coords

(ignored by partition_cv)

mode

Use future.apply::future_lapply() for parallelized execution if mode = "future", and lapply for sequential execution otherwise (mode = "sequential")

Details

Nearest-neighbour distances are calculated for each sample in the test set. These ⁠nrow(???$test)⁠ nearest-neighbour distances are then averaged. Aggregation methods other than mean can be chosen using the fun argument, which will be passed on to dataset_distance.

Value

A resampling or represampling object containing an additional. ⁠$distance⁠ component in each resampling object. The distance component is a single numeric value indicating, for each train / test pair, the (by default, mean) nearest-neighbour distance between the two sets.

See Also

dataset_distance represampling resampling

Examples

# Muenchow et al. (2012), see ?ecuador
nsp.parti <- partition_cv(ecuador)
sp.parti <- partition_kmeans(ecuador)
nsp.parti <- add.distance(nsp.parti, data = ecuador)
sp.parti <- add.distance(sp.parti, data = ecuador)
# non-spatial partioning: very small test-training distance:
nsp.parti[[1]][[1]]$distance
# spatial partitioning: more substantial distance, depending on number of
# folds etc.
sp.parti[[1]][[1]]$distance