pls (version 2.9-0)

cvsegments: Generate segments for cross-validation

Description

The function generates a list of segments for cross-validation. It can generate random, consecutive and interleaved segments, and supports keeping replicates in the same segment.

Usage

cvsegments(
  N,
  k,
  length.seg = ceiling(N/k),
  nrep = 1,
  type = c("random", "consecutive", "interleaved"),
  stratify = NULL
)

Value

A list of vectors. Each vector contains the indices for one segment. The attribute "incomplete" contains the number of incomplete segments, and the attribute "type" contains the type of segments.

Arguments

N

Integer. The number of rows in the data set.

k

Integer. The number of segments to return.

length.seg

Integer. The length of the segments. If given, it overrides k.

nrep

Integer. The number of (consecutive) rows that are replicates of the same object. Replicates will always be kept in the same segment.

type

One of "random", "consecutive" and "interleaved". The type of segments to generate. Default is "random".

stratify

Either a list of indices or an integer vector indicating which stratum each sample (or set of replicates) belongs to (see Details).

Author

Bjørn-Helge Mevik, Ron Wehrens and Kristian Hovde Liland

Details

If length.seg is specified, it is used to calculate the number of segments to generate. Otherwise k must be specified. If \(k*length.seg \ne N\), the \(k*length.seg - N\) last segments will contain only \(length.seg - 1\) indices.

If type is "random", the indices are allocated to segments in random order. If it is "consecutive", the first segment will contain the first \(length.seg\) indices, and so on. If type is "interleaved", the first segment will contain the indices \(1, length.seg+1, 2*lenght.seg+1, \ldots, (k-1)*length.seg+1\), and so on.

If \(nrep > \), it is assumed that each nrep consecutive rows are replicates (repeated measurements) of the same object, and care is taken that replicates are never put in different segments.

Warning: If k does not divide N, a specified length.seg does not divide N, or nrep does not divide length.seg, the number of segments and/or the segment length will be adjusted as needed. Warnings are printed for some of these cases, and one should always inspect the resulting segments to make sure they are as expected.

Stratification of samples is enabled by the stratify argument. This is useful if there are sub-groups in the data set that should have a proportional representation in the cross-validation segments or if the response is categorical (classifiation). If stratify is combined with nrep, stratify corresponds to the sets of replicates (see example).

See Also

Converting a factor to segments: fac2seg.

Examples

## Segments for 10-fold randomised cross-validation:
cvsegments(100, 10)

## Segments with four objects, taken consecutive:
(segs <- cvsegments(60, length.seg = 4, type = "cons"))
data(gasoline)
plsr(octane ~ NIR, data=gasoline, ncomp=5, validation="CV", segments=segs)

## Incomplete segments
segs <- cvsegments(50, length.seg = 3)
attr(segs, "incomplete")

## Leave-one-out cross-validation:
cvsegments(100, 100)
## Leave-one-out with variable/unknown data set size n:
n <- 50
cvsegments(n, length.seg = 1)

## Data set with replicates
cvsegments(100, 25, nrep = 2)
## Note that rows 1 and 2 are in the same segment, rows 3 and 4 in the
## same segment, and so on.

## Stratification
cvsegments(10, 3, type = "consecutive", stratify = c(rep(1,7), rep(2,3)))
## Note that the last three samples are spread across the segments
## according to the stratification vector.
cvsegments(20, 3, type = "consecutive", nrep = 2, stratify = c(rep(1,7), rep(2,3)))
## Note the length of stratify matching number of replicate sets, not samples.

## Converting a factor to segments
fac <- factor(c("a", "b", "a", "b", "c", "c"))
fac2seg(fac)
## Starting from a numeric vector
num <- c(1, 2, 1, 2, 3, 3)
fac2seg(factor(num))