Changelog • tv

Added an error to detect when parallel::mcMap() has an error in one of its forked processes.
Added parallelly::availableCores(omit = 1) as the new default for n_cores=.

tv::time_varying() gained a sort= argument, to warn the user that things ought to be sorted when x$datetime is a Date.
Added Peter Martin as a contributor, as he inspired the v2.0.0 rewrite.

The previous version of tv (< 2.0.0) accidentally silently dropped exposures that had zero (or negative) length. In 2.0.0 this became a non-informative error. It now has an informative error.

Completely overhauled the internals, for a decent speed boost. In dense data, the speed up in practice is on the order of ~2x. In sparse data, the speed up has been witnessed as high as ~25x.

Added return values to the documentation and resubmitted to CRAN.

Submit to CRAN.

Added logic to tv_min(), tv_max(), et al. to handle cases where there are relevant rows for the feature, but which are all missing. Usually you will want to filter these NA rows out a priori, because they’re almost never what you want.

Added grid.only= argument to time_varying().
Thank you, dplyr for changing your minds. Enforce dplyr (>= 1.1.1).
Added a “How does this actually work” section to the vignette.

Added multiple = 'all' to one dplyr::inner_join() and one dplyr::left_join().

Aggregations are now checked to make sure they exist. This includes, in particular, when the aggregation is NA.
Removed a few imports from the DESCRIPTION that weren’t being used.

Changed the default number of cores to check if "SLURM_CPUS_PER_TASK" is defined.

The exposures are now expected to be non-overlapping (although windows can touch when a start time equals an end time).

Reorganized the help pages.
Added special behavior for when "lookback_end" or "lookback_start" are NA.
The grid now includes "exposure_start" and "exposure_stop".
Added aggregation option "tv_median".
Added tests with testthat.

Updated vignette.
Forced warnings to be printed as they occur, instead of when tv::time_varying() finishes.

Added the aggregation option "sum".

Fixed an error message.

Added an attribute denoting what type of column each of the resulting columns is. This is useful for subsetting for, e.g., xgboost.

Added id= argument.

Initial port of code from kernds, removing all kds prefixes.