The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.
Removed usage of non-API STRING_PTR()
(#209).
R >=4.0.0 is now required, inline with tidyverse guidelines.
Updated snapshot tests to use the latest version of cli.
Fixed tests that failed under valgrind due to NA vs
NaN peculiarities (#198).
Bumped required versions of vctrs, cli, and rlang to their current CRAN versions.
Bumped required R version to >= 3.6.0.
Improved reported error calls and error messages throughout the package, and switched from glue to cli (#168, #188).
Index (.i) types that aren’t explicitly understood
by vctrs are now handled slightly better (#182).
The slide_index_*() and hop_index_*()
families now use vctrs::vec_rank() internally to compute a
dense rank, which should be a little faster than the previous home grown
approach (#177).
New slider_plus() and slider_minus()
developer facing helpers that allow package authors to register custom
double dispatch methods to override the default computation of
.i - .before and .i + .after when generating
index bounds internally. This is intended to allow the clock and almanac
packages to register methods so their custom types can be used natively
in slider (#91).
Removed ellipsis in favor of using the equivalent functions in rlang (#185).
Removed R_forceAndCall() fallback now that R
>=3.4.0 is required (#172).
Fixed -Wstrict-prototypes warnings as requested by
CRAN (#173).
Bumped minimal version of R to >=3.4.0 to align with tidyverse standards.
Bumped minimal version of rlang to >=1.0.6 and vctrs to >=0.5.0 (#165, #174).
Fixed redirecting URLs.
vec_order() to prepare for a
breaking change in vctrs (#153).Fixed a C alignment issue detected by CRAN’s USBAN machine
related to allocating vectors of long double.
Fixed a test that relied too strongly on the size of the C type,
long double, which can vary across platforms
(#147).
Fixed an out of sync vignette entry (#148).
New family of very fast specialized sliding functions:
slide_sum(), slide_index_sum(): for
rolling sums
slide_mean(), slide_index_mean(): for
rolling averages
slide_prod(), slide_index_prod(): for
rolling products
slide_min(), slide_index_min(): for
rolling minimums
slide_max(), slide_index_max(): for
rolling maximums
slide_any(), slide_index_any(): for
rolling any
slide_all(), slide_index_all(): for
rolling all
The slide_index_*() family now allows
.before and .after to be functions of 1
argument (the index) that compute the boundaries of the sliding window.
This can be extremely useful when the default, which computes
.i - .before and .i + .after, is not
applicable or correct for your needs. One use case is to set
.before = ~.x %m-% months(1) rather than
.before = months(1) to perform a 1 month rolling window in
a way that won’t generate NA values on invalid dates (like
1 month before 2019-03-31) (#139).
The slide_index_*() family has undergone some
internal changes to make it more compatible with custom vctrs classes
that could be provided as the index (.i), such as the
date-time classes in the clock package (#133, #130).
For the slide_index_*() family, it is now required
that .i - .before and .i + .after be castable
to .i by vctrs::vec_cast(). Similarly, for the
hop_index_*() family, .starts and
.stops must both be castable to .i
(#132).
New vignette, vignette("tsibble"), explaining how to
transition from tsibble to slider (#128).
vignette("rowwise") has been updated to use
cur_data() from dplyr 1.0.0, which makes it significantly
easier to do rolling operations on data frames (like rolling
regressions) using slider in a dplyr pipeline.
slide_period() and friends have slightly better
handling of size zero input when .complete = TRUE
(#111).
Better error messages for NA input with
.before, .after, .step and
.complete have been added (#110).
A few instances of possibly unsafe C protection usage have been fixed (#112).
Tests have been updated to use only numeric values in the
vctrs::new_date() constructor (#113).
As a followup to a change in slider 0.1.3, edge cases with size
zero input in hop() have also been fixed.
C code has been refactored to be less reliant on vctrs internals.
Updated to stay compatible with vctrs 0.3.0.
A few edge cases with size zero input in the index functions have been fixed.
The default for the .names_to argument of
*_dfr() variants has been updated to
rlang::zap() to match the default of the function it is
passed on to, vctrs::vec_rbind().
All *_vec() variants now maintain size stability
when auto-simplifying (i.e. when .ptype = NULL) (#78,
#93).
hop() and its variants no longer place the names of
.x on the output. Because there is no size
guarantee on the output, the size of .x can be different
than the size of the output, meaning that the names might not line up.
This also affects slide_period(), which is implemented
using a hop() variant (#75).
With data frames containing row names, slide() and
its variants now copy those row names onto the output. This is an
implicit benefit from vctrs gaining better support for data frame row
names.
NEWS.md file to track changes to the
package.These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.