Changelog • santoku

santoku 1.1.0

Core logic has been speeded up using raw pointers. This was vibe-coded by me and Claude Code. If it breaks, please file a bug report.
The experimental chop_spikes() and dissect() functions give common values of x their own singleton intervals.
On Unicode platforms, infinity will be represented as ∞ in breaks. Set options(santoku.infinity = "Inf") to use the old behaviour.
Singleton breaks are not labelled specially by default in chop_quantiles(..., raw = FALSE). This means that e.g. if the 10th and 20th percentiles are both the same number, the label will still be [10%, 20%].
When multiple quantiles are the same, santoku warns and returns the leftmost quantile interval. Before it would merge the intervals, creating labels that might be different to what the user asked for.
chop_quantiles() gains a recalc_probs argument. recalc_probs = TRUE recalculates probabilities using ecdf(x), which may give more accurate interval labels.
single = NULL has been documented explicitly in lbl_* functions.
Bugfix: brk_manual() no longer warns if close_end = TRUE (the default).

santoku 1.0.0

CRAN release: 2024-06-04

santoku is now considered stable.
chop_quantiles() and brk_quantiles() gain a new weights argument, letting you chop by weighted quantiles using Hmisc::wtd.quantile().
brk_quantiles() may now return singleton breaks, producing more accurate results when x has duplicate elements.
Some deprecated functions have been removed, and the raw argument to lbl_* functions now always gives a deprecation warning.

santoku 0.10.0

CRAN release: 2023-10-12

List arguments to fmt in lbl_* functions will be taken as arguments to base::format. This gives more flexibility in formatting, e.g., units breaks.
chop_n() gains a tail argument, to deal with a last interval containing less than n elements. Set tail = "merge" to merge it with the previous interval. This guarantees that all intervals contain at least n elements.
chop_equally() may return fewer than groups groups when there are duplicate elements. We now warn when this happens.
Bugfix: chop_n() could return intervals with fewer than n elements when there were duplicate elements. The new algorithm avoids this, but may be slower in this case.

santoku 0.9.1

CRAN release: 2023-03-08

endpoint_labels() methods gain an unused ... argument to satisfy R CMD CHECK.

santoku 0.9.0

CRAN release: 2022-11-01

Breaking changes

There are important changes to close_end.

close_end is now TRUE by default in chop() and fillet(). In previous versions:

chop(1:2, 1:2)
## [1] [1, 2) {2}   
## Levels: [1, 2) {2}

Whereas now:

chop(1:2, 1:2)
## [1] [1, 2] [1, 2]
## Levels: [1, 2]

close_end is now always applied after extend. For example, in previous versions:

chop(1:4, 2:3, close_end = TRUE)
## [1] [1, 2) [2, 3] [2, 3] (3, 4]
## Levels: [1, 2) [2, 3] (3, 4]

Whereas now:

chop(1:4, 2:3, close_end = TRUE)
## [1] [1, 2) [2, 3) [3, 4] [3, 4]
## Levels: [1, 2) [2, 3) [3, 4]

We changed this behaviour to be more in line with user expectations.

If breaks has names, they will be used as labels:
```
chop(1:5, c(Low = 1, Mid = 2, High = 4))
## [1] Low  Mid  Mid  High High
## Levels: Low Mid High  
```
Names can also be used for labels in probs in chop_quantiles() and proportions in chop_proportions().
There is a new raw parameter to chop(). This replaces the parameter raw in lbl_* functions, which is now soft-deprecated.
lbl_manual() is deprecated. Just use a vector argument to labels instead.
A labels argument to chop_quantiles() now needs to be explicitly named.

I expect these to be the last important breaking changes before we release version 1.0 and mark the package as “stable”. If they cause problems for you, please file an issue.

Other changes

New chop_fn(), brk_fn() and tab_fn() chop using an arbitrary function.
Added section on non-standard objects to vignette.

santoku 0.8.0

CRAN release: 2022-06-08

Breaking changes

lbl_endpoint() has been renamed to lbl_endpoints(). The old version will trigger a deprecation warning. lbl_endpoints() gains first, last and single arguments like other labelling functions.

Other changes

New chop_pretty(), brk_pretty() and tab_pretty() functions use base::pretty() to calculate attractive breakpoints. Thanks @davidhodge931.
New chop_proportions(), brk_proportions() and tab_proportions() functions chop x into proportions of its range.
chop_equally() now uses lbl_intervals(raw = TRUE) by default, bringing it into line with chop_evenly(), chop_width() and chop_n().
New lbl_midpoints() function labels breaks by their midpoints.
lbl_discrete() gains a single argument.
You can now chop ts, xts::xts and zoo::zoo objects.
chop() is more forgiving when mixing different types, e.g.:
- Date objects with POSIXct breaks, and vice versa
- bit64::integer64 and doubles
Bugfix: lbl_discrete() sometimes had ugly label formatting.

santoku 0.7.0

CRAN release: 2022-03-18

Breaking changes

In labelling functions, first and last arguments are now passed to glue::glue(). Variables l and r represent the left and right endpoints of the intervals.
chop_mean_sd() now takes a vector sds of standard deviations, rather than a single maximum number sd of standard deviations. Write e.g. chop_mean_sd(sds = 1:3) rather than chop_mean_sd(sd = 3). The sd argument is deprecated.
The groups argument to chop_evenly(), deprecated in 0.4.0, has been removed.
brk_left() and brk_right(), deprecated in 0.4.0, have been removed.
knife(), deprecated in 0.4.0, has been removed.
lbl_format(), questioning since 0.4.0, has been removed.
Arguments of lbl_dash() and lbl_intervals() have been reordered for consistency with other labelling functions.

Other changes

You can now chop many more types, including units from the units package, difftime objects, package_version objects, etc.
- Character vectors will be chopped by lexicographic order, with an optional warning.
- If you have problems chopping a vector type, file a bug report.
The glue package has become a hard dependency. It is used in many places to format labels.
There is a new lbl_glue() function using the glue package. Thanks to @dpprdan.
You can now set labels = NULL to return integer codes.
Arguments first, last and single can be used in lbl_intervals() and lbl_dash(), to override the first and last interval labels, or to label singleton intervals.
lbl_dash() and lbl_discrete() use unicode em-dash where possible.
brk_default() throws an error if breaks are not sorted.

Bugfixes

Bugfix: tab() and friends no longer display an x as the variable name.
Bugfix: lbl_endpoint() was erroring for some types of breaks.

santoku 0.6.0

CRAN release: 2021-11-04

New arguments first and last in lbl_dash() and lbl_discrete() allow you to override the first and last interval labels.
Fixes for CRAN.

santoku 0.5.0

CRAN release: 2020-08-27

Negative numbers can be used in chop_width().
- This sets left = FALSE by default.
- Also works for negative time intervals.

santoku 0.4.1

CRAN release: 2020-06-16

Bugfix: chop(1:4, 1) was erroring.

santoku 0.4.0

CRAN release: 2020-06-09

Interface changes

The new version has some interface changes. These are based on user experience, and are designed to make using chop() more intuitive and predictable.

chop() has two new arguments, left and close_end.
- Using left = FALSE is simpler and more intuitive than wrapping breaks in brk_right().
- brk_left() and brk_right() have been kept for now, but cannot be used to wrap other break functions.
- Using close_end is simpler than passing close_end into brk_left() or brk_right() (which no longer accept this argument directly).
- left = TRUE by default, except for non-numeric objects in chop_quantiles() and chop_equally(), where left = FALSE works better.
close_end is now FALSE by default.
- This prevents user surprises when e.g. chop(3, 1:3) puts 3 into a different category than chop(3, 1:4).
- close_end is TRUE by default for chop_quantiles(), chop_n() and similar functions. This ensures that e.g. chop_quantiles(x, c(0, 1/3, 2/3, 1)) does what you would expect.
The groups argument to chop_evenly() has been renamed from groups to intervals. This should make it easier to remember the difference between chop_evenly() and chop_equally(). (Chop evenly into n equal-width intervals, or chop equally into n equal-sized groups.)
knife() has been deprecated to keep the interface slim and focused. Use purrr::partial() instead.

Other changes

Date and datetime (POSIXct) objects can now be chopped.
- chop_width() accepts difftime, lubridate::period or lubridate::duration objects
- all other chop_ functions work as well.
Many labelling functions have a new fmt argument. This can be a string interpreted by sprintf() or format(), or a 1-argument formatting function for break endpoints, e.g. scales::label_percent().
Experimental: lbl_discrete() for discrete data such as integers or (most) dates.
There is a new lbl_endpoint() function for labelling intervals solely by their left or right endpoint.
brk_mean_sd() now accepts non-integer positive numbers.
Add brk_equally() for symmetry with chop_equally().
Minor tweaks to chop_deciles().
Bugfix: lbl_format() wasn’t accepting numeric formats, even when raw = TRUE. Thanks to Sharla Gelfand.

santoku 0.3.0

CRAN release: 2020-01-24

First CRAN release.
Changed kut() to kiru(). kiru() is an alternative spelling for chop(), for use when the tidyr package is loaded.
lbl_sequence() has become lbl_manual().
lbl_letters() and friends have been replaced by lbl_seq():
- to replace lbl_letters() use lbl_seq()
- to replace lbl_LETTERS() use lbl_seq("A")
- to replace lbl_roman() use lbl_seq("i")
- to replace lbl_ROMAN() use lbl_seq("I")
- to replace lbl_numerals() use lbl_seq("1")
- for more complex formatting use e.g. lbl_seq("A:"), lbl_seq("(i)")

santoku 0.2.0

Added a NEWS.md file to track changes to the package.
Default labels when extend = NULL have changed, from [-Inf, ... and ..., Inf] to [min(x), ... and ..., max(x)].