Experimental: When calling
incidence() if users specify an integer(ish) interval on an
(integerish) date input these inputs will be coerced in to
grates_int_period objects. This fixes
https://github.com/reconverse/incidence2/issues/121 and (broadly)
restores earlier behaviour of the incidence() function.
This functionality is subject to change and, for that reason, is not
widely documented.
library(incidence2)
x <- data.frame(onset = c(1, 3, 4, 4, 3, 4, 10))
# 2.5.0 release would error, e.g.
incidence(x, "onset", interval = 1)
#> Error in as_period.default(): Not implemented for class [numeric].
# This release - now works
incidence(x, "onset", interval = 1)
#> # incidence: 4 x 3
#> # count vars: onset
#> date_index count_variable count
#> <intper> <chr> <int>
#> 1 1 onset 1
#> 2 3 onset 2
#> 3 4 onset 3
#> 4 10 onset 1incidence() gains a fill parameter.
This is passed to complete_dates() so is only valid when
the argument complete_dates = TRUE.
For convenience we now reexport select() from dplyr as well as
unnest() and unpack() from tidyr.
Alt text has been added to all vignette images.
Improved input checking for keep_first() and
keep_last().
Temporarily remove the ympes dependency due to up stream changes.
No other user facing changes.
The bootstrap_incidence(),
estimate_peak() and first_peak() functions,
and have been integrated from the downstream i2extras
package.
Added group-aware methods for common generics in base R, dplyr and tidyr:
split.incidence2()mutate.incidence2()summarise.incidence2()nest.incidence2()Added coercion methods for tibble and data.table:
as_tibble.incidence2()as.data.table.incidence2()New methods, $<-.incidence2 and
rbind.incidence2, that drop the incidence2 class if the
required invariants are broken.
New functions incidence_() and
regroup_() that work similar to their existing namesakes
save for additional support for tidy-select
semantics in some of their arguments.
Named vectors can now be used for the groups and
counts arguments of incidence() to allow
renaming prior to aggregation. Previously this was only possible with
the date_index input.
incidence() gains a complete_dates
argument defaulting to FALSE. If set this is equivalent of
a call to incidence() followed by a call to
complete_dates() with the default arguments. Users wanting
more flexibility can still call the complete_dates()
function with different arguments.
All of the aforementioned new methods would previously have
dispatched on the underlying data.frame method. If you were relying on
this behaviour then you will now need to add a call to
as.data.frame() prior to continuing your pipeline.
incidence2 objects are now built upon tibbles rather
than standard data frames. This means where we do not provide methods
for incidence2 objects tibble (as opposed to data.frame)
methods will be called. An overview of the differences between tibbles
and data.frames can be found at
https://tibble.tidyverse.org/articles/invariants.html.
incidence() now warns if a count variable contains
missing values and encourages users to handle these prior to calling
incidence().
The Package now has a hard dependency on the R version (>= 4.1.0).
plot.incidence2() now works again when applied to
incidence2 objects with a grates_period
date_index. This was inadvertently broken in the 2.2.1
release. Thanks to @Bisaloo for the report (#110).plot.incidence2() gains arguments
n_breaks, fill, show_cases and
legend allowing for a wider range of plot styles. See
vignette("incidence2", package = "incidence2") for
examples.interval = "day" or
interval = daily in a call to incidence will force the
resultant date_index variable to be a Date.
Functionally this is a wrapper around as.Date() that
ensures the underlying data are whole numbers and that time zones are
respected.incidence() will now warn if objects are created
with POSIXct columns. The motivation for this is that,
internally, POSIXct objects are represented as seconds
since the UNIX epoch and, in our experience, this level of granularity
is not normally desired for aggregation.
The by parameter of complete_dates() is
now defunct. This was previously passed to an underlying
seq function when, in practice, it should always have been
forced to 1 to match the precision of the underlying
date_index.
complete_dates() will now error if called on an
input with a allow_POSIXct = TRUE to maintain old
behaviour.
Version 2.0.0 is a major, breaking release of incidence2. We have undertaken a significant refactor that both simplifies the underlying code base and makes the user interface more consistent and robust. Although the main changes are highlighted below, users are strongly advised to read through the accompanying documentation and vignettes.
We no longer support NSE (e.g. tidyselect semantics) within the package. Our motivation for removing support for NSE are both the complexity it can bring to the underlying code (making long term maintenance harder) and the complexity it can cause for other users / developers who want to build on top of incidence2.
new_incidence(), validate_incidence(),
build_incidence(), get_n(),
get_interval(), get_timespan() and
facet_plot() are now defunct and will error if
called.
complete_counts() is now renamed
complete_dates() and gains two new parameters,
expand and by. If expand is TRUE
(default) then complete_dates() will attempt to use
function(x) seq(min(x), max(x), by = by) to generate a
complete sequence of dates.
The incidence() function now always returns output
in long format with dedicated columns for the count variables and values
(set by arguments count_names_to and
count_values_to).
incidence() is now less flexible in what it can
accept for the interval argument. For more complex date
groupings users are encouraged to perform their require data
transformations prior to calling incidence().
The default plotting of incidence objects as been greatly simplified. Sensible defaults have been chosen to enable a quick visual overview of incidence objects but users are advised to call ggplot2 directly for more bespoke plotting.
cumulate() functionality (previously
deprecated in 1.2.0).fill argument in complete_counts()
is now 0 rather than NA.incidence() when more than one column was
given for the date_index.new_incidence(): A minimal incidence constructor.validate_incidence(): Check for internal consistency of
incidence-like object.build_incidence(): Allows you to construct an incidence
object whilst specifying your own date grouping function.format.incidence()cumulate() will now give a deprecation error. We have
removed the function to avoid users erroneously regressing against a
cumulative count.incidence() when dates were a character
vector and the the default, daily, interval was specified.dplyr to handle list based columns
(e.g. record-type objects from vctrs). For data.frames with
only atomic columns, data.table is still used.complete_counts().plot() and facet_plot() now have a
centre_dates argument which can be set to
FALSE to get histogram-esque date labels for single month,
quarter and yearweek groupings.Due to multiple changes in the underlying representation of incidence2 objects this release may possibly break old workflows particularly those relying on the old implementations of date grouping:
grates for
date grouping. This introduces the s3 classes yrwk,
yrmon, yrqtr, yr,
period and int_period as well as associated
constructors which incidence now builds upon. As a result
of this the aweek
dependency has been dropped.keep_first and keep_last
functions.incidence objects now faster due to
underlying use of data.table.show_cases = TRUE (see #42).count variable of a
pre-aggregated input to incidence function.