Randomly Break Ties in Data
This is a generic function intended to randomly break tied data in a
way similar to what
jitter does: tie-breaking is
performed by shifting all data points by a random amount.
The surveillance package defines methods for matrices,
"epidataCS", and a default method for numeric vectors.
untie(x, amount, ...) # S3 method for epidataCS untie(x, amount = list(t=NULL, s=NULL), minsep = list(t=0, s=0), direction = "left", keep.sources = FALSE, ..., verbose = FALSE) # S3 method for matrix untie(x, amount = NULL, minsep = 0, constraint = NULL, giveup = 1000, ...) # S3 method for default untie(x, amount = NULL, minsep = 0, direction = c("symmetric", "left", "right"), sort = NULL, giveup = 1000, ...)
the data to be untied.
upper bound for the random amount by which data are shifted.
NULLmeans to use a data-driven default, which equals the minimum separation of the data points for the non-symmetric default method and its half for the symmetric default method and the
minimum separation of jittered points. Can only be obeyed if much smaller than
amount(also depending on the number of points).
minsep>0is currently only implemented for the spatial (matrix) method.
TRUE, the original list of possible event sources in
x$events$.sourceswill be preserved. For instance, events observed at the same time did by definition not trigger each other; however, after random tie-breaking one event will precede the other and considered as a potential source of infection for the latter, although it could just as well be the other way round. Enabling
keep.sourceswill use the
.sourceslist from the original (tied)
"epidataCS"object. Note, however, that an update is forced within
twinstimif a subset of the data is selected for model fitting or if a different
an object of class
"SpatialPolygons"representing the domain which the points of the matrix should belong to -- before and after jittering.
number of attempts after which the algorithm should stop trying to generate new points.
"right", indicating in which direction vector elements should be shifted.
logical indicating if the jittered vector should be sorted. Defaults to doing so if the original vector was already sorted.
"epidataCS"-method: arguments passed to the
giveup). Unused in other methods.
logical passed to
For numeric vectors (default method), the jittered version is the
same as for
jitter(x, amount=amount), if
amount is non-
and otherwise uses
runif(length(x), 0, amount).
For matrices, a vector uniformly drawn from the disc with radius
amount is added to each point (row).
amount is a list stating the amounts
for the temporal and/or spatial dimension, respectively. It then
uses the specific methods with arguments
sort=TRUE. Note that this implements a
simplistic approach of tie-breaking where all events are assumed to be
subject to the same amounts of censoring, and the default amounts may
not be sensible choices.
# vector example set.seed(123) untie(c(rep(1,3), rep(1.2, 4), rep(3,3)), direction="left", sort=FALSE) # spatial example data(imdepi) coords <- coordinates(imdepi$events) table(duplicated(coords)) plot(coords, cex=sqrt(multiplicity(coords))) set.seed(1) coords_untied <- untie(coords) stopifnot(!anyDuplicated(coords_untied)) points(coords_untied, col=2) # shifted by very small amount in this case