dclone: Cloning R objects

In datacloning/dclone: Data Cloning and MCMC Tools for Maximum Likelihood Methods

Cloning R objects

Description

Makes clones of R objects, that is values in the object are repeated n times, leaving the original structure of the object intact (in most of the cases).

Usage

dclone(x, n.clones=1, ...)
## Default S3 method:
dclone(x, n.clones = 1, attrib=TRUE, ...)
## S3 method for class 'dcdim'
dclone(x, n.clones = 1, attrib=TRUE, ...)
## S3 method for class 'dciid'
dclone(x, n.clones = 1, attrib=TRUE, ...)
## S3 method for class 'dctr'
dclone(x, n.clones = 1, attrib=TRUE, ...)
## S3 method for class 'list'
dclone(x, n.clones = 1,
    multiply = NULL, unchanged = NULL, attrib=TRUE, ...)
## S3 method for class 'environment'
dclone(x, n.clones = 1,
    multiply = NULL, unchanged = NULL, attrib=TRUE, ...)
dcdim(x, drop = TRUE, perm = NULL)
dciid(x, iid=character(0))
dctr(x)

Arguments

x	An R object to be cloned, or a cloned object to print.
n.clones	Number of clones.
multiply	Numeric or character index for list element(s) to be multiplied by n.clones instead of repetitions (as done by dclone.default).
unchanged	Numeric or character index for list element(s) to be left unchanged.
attrib	Logical, TRUE if attributes are to be attached.
drop	Logical, if TRUE, deletes the last dimension of an array if that have only one level.
perm	The subscript permutation value, if the cloning dimension is not the last.
iid	Character (or optionally numeric or logical). Column(s) to be treated as i.i.d. observations. Ignored when x is a vector.
...	Other arguments passed to function.

Details

dclone is a generic function for cloning objects. It is separate from rep, because there are different ways of cloning, depending on the BUGS code implementation:

(1) Unchanged: no cloning at all (for e.g. constants).

(2) Repeat: this is the most often used cloning method, repeating the observations row-wise as if there were more samples. The dctr option allows repeating the data column-wise.

(3) Multiply: sometimes it is enough to multiply the numbers (e.g. for Binomial distribution).

(4) Add dimension: under specific circumstances, it is easier to add another dimension for clones, but otherwise repeat the observations (e.g. in case of time series, or for addressing special indexing conventions in the BUGS code, see examples dcdim and dclone.dcdim).

(5) Repeat pattern (i.i.d.): this is useful for example when a grouping variable is considered, and more i.i.d. groups are to be added to the data set. E.g. c(1, 1, 2, 2) is to be cloned as c(1, 1, 2, 2, 3, 3, 4, 4) instead of c(1, 1, 2, 2, 1, 1, 2, 2).

Value

An object with class attributes "dclone" plus the original one(s). Dimensions of the original object might change according to n.clones. The function tries to take care of names, sometimes replacing those with the combination of the original names and an integer for number of clones.

dcdim sets the class attribute of an object to "dcdim", thus dclone will clone the object by adding an extra dimension for the clones.

dciid sets the class attribute of an object to "dciid", thus dclone will clone the object by treating columns defined by the iid argument as i.i.d. observations. These columns must be numeric. This aims to facilitates working with the INLA package to generate approximate marginals based on DC. Columns specified by iid will be replaced by an increasing sequence of values respecting possible grouping structure (see Examples).

Lists (i.e. BUGS data objects) are handled differently to enable element specific determination of the mode of cloning. This can be done via the unchanged and multiply arguments, or by setting the behaviour by the dcdim function.

Environments are coerced into a list, and return value is identical to dclone(as.list(x), ...).

Author(s)

Peter Solymos, implementation is based on many discussions with Khurram Nadeem and Subhash Lele.

References

Lele, S.R., B. Dennis and F. Lutscher, 2007. Data cloning: easy maximum likelihood estimation for complex ecological models using Bayesian Markov chain Monte Carlo methods. Ecology Letters 10, 551–563.

Lele, S. R., K. Nadeem and B. Schmuland, 2010. Estimability and likelihood inference for generalized linear mixed models using data cloning. Journal of the American Statistical Association 105, 1617–1625.

Solymos, P., 2010. dclone: Data Cloning in R. The R Journal 2(2), 29–37. URL: https://journal.r-project.org/archive/2010-2/RJournal_2010-2_Solymos.pdf

Examples

## scalar
dclone(4, 2)
## vector
(x <- 1:6)
dclone(x, 2)
## matrix
(m <- matrix(x, 2, 3))
dclone(m, 2)
## data frame
(dfr <- as.data.frame(t(m)))
dclone(dfr, 2)
## list
(l <- list(n = 10, y = 1:10, x = 1:10, p = 1))
dclone(l, 2)
dclone(as.environment(l), 2)
dclone(l, 2, attrib = FALSE)
dclone(l, 2, multiply = "n", unchanged = "p")
## effect of dcdim
l$y <- dcdim(l$y)
dclone(l, 2, multiply = "n", unchanged = "p")
## time series like usage of dcdim
z <- data.matrix(rnorm(10))
dclone(dcdim(z), 2)
## usage if dciid
ll <- dciid(data.frame(x=1:10, y=1:10), iid="y")
dclone(ll, 2)
## respecting grouping structure in iid
ll$y <- rep(1:5, each=2)
(dci <- dclone(ll, 2))
nclones(dci)
## repeating the data column-wise
dclone(dctr(m), 2)

datacloning/dclone documentation built on Sept. 29, 2024, 3:21 p.m.