Scientific notation on the y axis (e.g. y.scientific = TRUE) failed with ggplot2 4.0.0 with the error ‘invalid mathematical annotation’
Calculations in titles (title, subtitle, caption, tag, x.title, y.title and category.title) now always use the columns of the data first. Before, an object with the same name in the calling environment took precedence, so that e.g. caption = min(format(date, "%Y")) used the function date() instead of the column date, and a global object age was used instead of the column age.
In maps (sf objects), colour_fill now also sets the colours of a discrete category; before, the colours of ggplot2 were always used.
The percentage breaks of density plots now follow the computed density. Before, they were based on the values of x, which could give a single, odd break such as 99%.
The vignettes were rewritten: ‘Introduction to plot2’ (vignette("plot2")) and ‘Plot types’ (vignette("supported_types")). The package website has new articles on colours and themes, maps, fonts, and plot2() for ggplot2 users. The examples in ?plot2 were reduced to the core usage.
In plot2() for sf objects, the arguments crs and datalabels.centroid now come after ..., so they must be named (e.g. crs = 28992). This makes the method consistent with the generic, as R CMD check requires.
Long rotated datalabels no longer run beyond the plot: unless y.expand or y.limits is set, the value axis gets extra room for them, estimated for a typical figure size.
The whiskers of boxplots with a category now get their colours from the colour scale, which removes an internal workaround that repeated colour vectors.
With facet.drop = FALSE, empty panels no longer cause R warnings when drawing: bars without a category are no longer dodged, and a discrete x axis is shared between the panels (with a message).
Rotated x-axis labels (x.label_angle) are now anchored at their ticks for every angle, also in horizontal plots, using ggplot2::guide_axis(angle = ); setting x.label_align still uses your own alignment.
In horizontal plots, the values now always read from top to bottom in the order of x.sort, as they read from left to right in vertical plots. Previously, factors (such as age groups), x.sort = "alpha" and x.sort = FALSE read in reverse.
Rotated datalabels (any datalabels.angle) and the datalabels of horizontal plots are now anchored at the end of their bar with the same small gap for every label; before, the gap grew with the length of the label, and at 270 degrees the labels overlapped the bars.
size, width and linewidth now give a clear error when they are not a non-negative number, instead of failing only when the plot is printed.
type with more than one value now warns that only the first is used, instead of silently using it.
In back-to-back plots, each category has the same colour on both sides, also when a category is missing on one side, and the legend is shown once.
Back-to-back plots now draw both halves on the same value scale, so equal bar lengths mean equal values; a user-set y.limits is kept.
In x.max_txt and friends, %p is now the share of the total of y in the rest item (it was the share of rows); the documentation now states that x.max_items includes the rest item and that %n is the number of merged values
create_interactively() has a new argument run: run = FALSE returns the app object, e.g. for deployment on a server; see the new section ‘Deployment’ in ?create_interactively
create_interactively() refuses exports above 40 megapixels, and checks export sizes on the server
The copy button of create_interactively() copies in the browser, so it also works when the app runs on a server; the app no longer needs the clipr and fansi packages
create_interactively() shows plot2’s messages also on a server, and shows the actual error message
create_interactively() has inputs for the secondary axis
create_interactively(): vectors such as y.limits = c(0, 10), formulas and label functions can be entered, and text that looks like a number stays text
create_interactively() refers to passed data by its name in the generated code, handles data frames with non-syntactic names, and offers data frames from the global environment only in an interactive session
create_interactively() no longer runs text from the browser as R code: the plot2() call is built from checked values, and evaluated apart from the global environment
Uploading in create_interactively(): the encoding is chosen from a list of valid values, and .txt files are read with the chosen delimiter
create_interactively() no longer accepts R serialisation files (.rds, .rda) for upload, since reading them can run code; pass such data with the data argument. With upload_tab = FALSE (the default) the app offers no way to upload files at all
create_interactively() serves only a copy of the logo, no longer the whole folder that contains it
Column names are no longer parsed as R code in back-to-back and UpSet plots, so data with unusual column names (such as my col) work, and column names can never run code
?plot2 has a new section ‘Automatic choices’ that lists every decision plot2 makes by itself, its rule and the argument that overrides it
Much faster on large data: datalabels are formatted only for the rows that are drawn, e.g. 200,000 rows now take about 0.2 s instead of 1.5 to 3 s
Arguments renamed to follow one naming rule (see ‘Argument names’ in ?plot2): colour_opacity is now colour_transparency (and get_colour(opacity) is transparency), x.lbl_angle, x.lbl_align and x.lbl_italic are x.label_*, facet.repeat_lbls_* are facet.repeat_labels_*, facet.fill is facet.colour_fill, sankey.* are sankey_*, and the options dec_mark and big_mark are plot2.decimal.mark and plot2.big.mark. The old argument names give an error that names the new one
New: legend = FALSE hides the legend; stacked = "fill" stacks to 100%
x.mic was removed; add + AMR::scale_x_mic() to the plot instead. x.lbl_taxonomy was removed; use x.labels = AMR::labels_taxonomy instead. The AMR package is no longer a suggested dependency
Giving both category.title and legend.title is now an error, since they are the same setting
Automatic defaults are now NULL in the usage, and explicit values mean what they say: e.g. datalabels = TRUE always shows all labels, while the default NULL shows at most 25; summarise_function = FALSE (no longer NULL) turns off summarising; an explicit y.percent_break is kept as given
plot2() on a matrix now says that x, y and category are ignored, instead of ignoring them silently
In functions and loops, default titles show the column name instead of code such as {{ col }} or .data[[v]]
The arguments of plot2() are now defined once; the methods take them from the generic, so they can no longer drift apart
Spider plots (type = "spider") now support facets, and keep their spider shape when datalabels are shown together with a title or facets
plot2() on a plain numeric matrix now plots all columns (as categories) instead of only the first, and uses the names of named dimensions as titles
x.limits, y.limits and category.limits must now be of length 2 (use NA for a limit that follows the data); a single value previously gave a zero-width axis without any data
Numeric category scales now use the data range rounded to pretty values (e.g. 0 to 100 for values from 12 to 87) instead of an undocumented heuristic; with three colours and category.midpoint, the scale is symmetric around the midpoint
add_sf() now uses datalabels.colour for its labels, works with any name of the geometry column, and no longer shifts labels by default (datalabels.nudge_y = 0; the former 2500 only suited the Dutch RD New projection)
An invalid colour in colour now gives a warning and is replaced by grey on its own, keeping the other colours and their names (previously the whole vector became one grey); register_colour() gives an error for invalid colours, and registered colour names are matched case-insensitively
add_type() now adds the layer when data is given to a plot without top-level data, warns when there is no data at all (instead of silently returning the plot unchanged), and supports beeswarm types
Plot types without dedicated support (such as "step") no longer receive arguments they do not accept, so ggplot2 no longer warns about unknown parameters; the notice that such a type is ‘only loosely supported’ is now a message instead of a warning
text_factor now scales every text element by exactly that factor, also in themes with relative sizes such as theme_bw()
A partial theme, such as theme = theme(panel.background = element_blank()), now works: it is added on top of theme_minimal2() and its elements take precedence
Axis labels now switch to scientific notation automatically only from one billion (1e9) or for values below 0.0001; large counts such as 2 500 000 are shown with big marks, and small values get more decimals instead of scientific notation
With facets, the automatic choice between free and fixed y (and x) scales is now always reported in a message and documented, since it depends on the data
legend.position now also accepts FALSE (no legend), "inside", and two numbers between 0 and 1 to place the legend inside the plot, e.g. legend.position = c(0.95, 0.95)
The secondary series of a column plot (y_secondary) is now a line with points by default, instead of a second set of bars drawn over the first
Secondary y axis: the secondary series is now actually drawn on the scale of its axis (previously it could be drawn with its original values), an all-zero series no longer errors, and negative values are mapped correctly
New argument datalabels.percent_by: "%p" datalabels are now percentages within each x value (so stacked bars sum to 100% per bar), or within each category, facet, or the total; previously they were always relative to the grand total of the plot. Without a category, the default gives percentages within each facet
Numeric x values that look like years (2000 to 2050) or months (1 to 12) now stay numeric for continuous types such as "point" and "line", and are only treated as years or months when they are whole numbers; columns of years without type work as before
An unknown or ambiguous type now gives an error with suggestions (“did you mean ‘boxplot’?”) instead of silently using the first geom whose name contains the text; abbreviations such as "hist" must be the start of a geom name
plot2("name") now looks up a data frame of that name only in the calling environment, not along the search path or in plot2’s namespace; plot2(numbers, y = values) now uses the vector as x
Passing data = to plot2() with anything other than a formula now gives a clear error instead of ‘argument matches multiple formal arguments’
The formula interface now accepts computed terms such as log(mpg) ~ hp or I(hp + 1), a facet after | (y ~ x + category | facet), and works after a pipe: mtcars |> plot2(mpg ~ wt)
Titles and category values with apostrophes (“Children’s E. coli”) no longer fail with markdown, a single dollar sign (“Costs in $") or currency amounts no longer turn a title into plotmath (plotmath needs `$…$` as in Pandoc), and markdown is no longer detected across separate titles
Column names and values containing regular expression characters (such as rate (%) or n[1) no longer break plot2(), and case-insensitive matching no longer changes the meaning of patterns such as \D
Dates in legends and on the x axis now follow the system locale instead of Dutch; the new option plot2.language ("en" or "nl") sets the language of month names and of the words plot2 generates (automatic titles, y.24h, y.age), which no longer depend on the machine’s locale. Changes of the time locale are always restored
Date formats such as x.date_labels = "Week ww" keep words intact (previously “%Veek”), text in single quotes is kept literally, quarters (q, qq) also work on the x axis, month formats return text like all other formats, and time zones are kept when years are unified
lubridate is no longer needed
Custom fonts: Google Fonts are now downloaded into plot2’s user cache folder (tools::R_user_dir("plot2", "cache")) instead of into the installed showtextdb package, and fonts not used for 90 days are removed from it. A custom font no longer switches on showtext for the whole session: it is used only while that plot is drawn, so other plots are unaffected. The packages showtext, showtextdb and sysfonts moved to Suggests and are needed only for custom fonts
Fewer hard dependencies: broom and cleaner moved to Suggests (broom is needed only to plot model objects that ggplot2::fortify() cannot handle) and crayon was dropped in favour of cli
cli, rlang, scales, grid and grDevices are now listed in Imports, as plot2 always uses them; stringr is no longer needed for sorting
Messages and warnings of plot2() are now R conditions. Messages (class plot2_message, and plot2_caution for cautions) are still printed only in interactive sessions, now grouped at the end of the call in the order they occurred, and can be caught or hidden with suppressMessages(); options(plot2.silent = TRUE) now works. Warnings (such as “No observations” or values that could not be plotted) are now R warnings of class plot2_warning, also in scripts, R Markdown and Quarto
Fixed: internal state of one plot2() call could leak into the next call, or be wiped by a nested call (back-to-back and UpSet plots), which could give wrong axis titles and messages that belonged to an earlier plot
x.zoom (and zoom) no longer sets numeric limits on a discrete x axis
x.zoom, y.zoom and zoom now zoom to the data with ggplot2’s usual 5% expansion (previously 50% for x and 25% for y), zoom = TRUE is identical to x.zoom = TRUE, y.zoom = TRUE, and it no longer expands beyond y.limits
The help pages of legend.position and of the package options (?plot2-options) now state the actual defaults
Main vignette: corrected several explanations that did not match their example code
README: corrected the installation instructions (not on CRAN yet) and two examples that failed or did not do what the text said
type = "beeswarm" without the ggbeeswarm package now gives an install hint instead of ‘geom_beeswarm() does not exist’, and all type aliases and sorting values are documented
Corrected typos and factual errors in the help pages of plot2(), add_type(), md_to_expression(), the package options and the netherlands data set
The vignettes build when suggested packages (ggplot2movies, ggbeeswarm, ggforce, sf, patchwork, viridisLite) are not installed; the affected examples are then skipped.
CI now also checks R 4.1 (the declared minimum) and a set-up with only the hard dependencies
x.complete, category.complete and facet.complete now work for character and factor values (crossed with the other directions), and complete date-times at the observed period instead of per second
The colour palette "ggplot2" (the default) now gives exactly the colours ggplot2 itself uses, also for more than one colour
The help page of get_colour() no longer changes at every documentation run
Scientific axis labels now use a multiplication sign (×) that renders on every device, instead of a console bullet; euros() and dollars() no longer use scientific notation for large values
Violin plots no longer fail when groups have a single observation
Without y and without a numeric column, plot2() now counts rows (y = n()) instead of stopping; Sankey plots can be drawn from raw rows
type = "barpercent" without y now counts rows instead of summing the first numeric column
options(plot2.default_type = ) now accepts the same values as type, such as "line" and "l"
colour_opacity now also applies to the default colours and to all built-in R palettes
Subsetting a colour vector with [ or [[ now keeps the colour class
theme_minimal2() text sizes are now plain numbers instead of grid units
theme_minimal2(...) now changes only the given properties, like + theme(...)
coord_spider() radial labels now follow start and the theme’s font family and colour
add_smooth(legend.value = ) now works on non-date x axes
move_layer() with a positive move no longer duplicates or drops layers
add_errorbar() now works for plots without a category, and with facets
plot2() on sf data with x or y now gives the intended caution instead of an error
y_secondary.colour and y_secondary.colour_fill now apply also when colour is not set
plot2(df, y = col) no longer fails when the data have no other column to use as x
datalabels given with an explicit point or line type are no longer silently dropped
Numeric x and y with identical values now give a scatter plot instead of a histogram; a histogram is only chosen when x and y are the same column
UpSet plots no longer use a soft-deprecated tidyselect idiom
Box plots no longer use the fatten argument, which is deprecated in ggplot2 4.0.0
The returned plot data no longer contain internal columns starting with _; datalabels and Sankey helper columns get readable names (datalabels, sankey_x, …)
Character and factor vectors given as a long expression, or named y, can now be plotted
Sankey plots without two x variables or without y now give a clear error instead of a cryptic one, and UpSet plots accept a manual x.sort vector
Density plots now get Density as y axis title instead of the name of the x variable
category.focus now also works with the default colours; the focused category keeps its own colour
Long titles with accented, Greek, Cyrillic or CJK characters are now wrapped like other titles
x.limits now works for a discrete x axis: x.limits = c("b", "a") shows only these items, in this order
x.lbl_angle and x.lbl_align now also apply to horizontal plots, where the x axis labels are on the vertical axis
summarise_function = NULL now keeps all values instead of failing for duplicate x values, as documented
summarise_function must now return a single value per group, with a clear error otherwise; groups whose summary is NA are no longer dropped silently
A column named n in x, category or facet no longer breaks the summarising of y
Fixed: back-to-back plots with small counts failed, and y = c(2, 3) was read as column positions
Fixed: back-to-back plots failed when the facet was given as an expression, such as facet = tolower(sex)
Fixed: a date-time (POSIXct) category produced thousands of nonsense legend breaks; a category with only a few dates now shows one legend entry per date
Fixed: a factor y with data labels failed with an nchar() error
Fixed: a category or datalabels column of only 0s or 1s was mistaken for FALSE/TRUE, and columns with missing values crashed plot2()
Fixed: area plots with a category were shifted sideways; unstacked areas are now overlaid at their x values
Fixed: giving fewer colours than categories failed with “Insufficient values in manual scale”; the colours are now recycled with a warning
Fixed: a histogram of constant values was empty because the automatic binwidth was 0
Fixed: y.transform = "reverse" applied y.expand to the wrong sides and in the wrong units
Fixed: category.percent = TRUE failed for numeric categories
Fixed: a logarithmic category.transform failed when the automatic colour scale started at 0
Fixed: transformation objects such as scales::transform_log10() in x.transform, y.transform and category.transform caused errors
Fixed: x.mic = TRUE failed with current versions of the AMR package
Fixed: logarithmic and other transformed x axes got linear breaks
Fixed: horizontal = TRUE with a numeric x removed all data points
Fixed: x.limits on a numeric x announced x.expand = 0 but still padded the axis, so it did not end at the given limits
Fixed: an explicit 0 in y.limits was silently replaced by the lowest (negative) data value
Fixed: percentage axes (and age and 24-hour axes) with negative values had breaks at odd values such as -28%, -18%, -8%
Fixed: axes and colour bars for small whole numbers had no breaks below zero
Error messages no longer show internal function calls
Fixed: several short hex colours (such as "#f0a") were all replaced by the first one, and invalid hex codes could pass validation
Fixed: an invalid dec_mark or big_mark option no longer becomes the decimal or big mark; it gives an R warning and falls back to the default
Fixed: UpSet plots with print = TRUE raised an error after printing; print = TRUE now always returns the plot invisibly
Fixed: types that merely contain “bar” (such as "errorbar" and "crossbar") were turned into horizontal columns
Fixed: theme = theme_bw (a theme function instead of its name) failed in several plot types
Fixed: plot2() no longer changes the random-number stream (.Random.seed); the tip about options(plot2.silent = TRUE) is now shown once per session
Fixed: the message about rows removed with na.rm = TRUE was garbled (“rowss”)
Fixed: date-time (POSIXct) columns could be chosen as the automatic y variable, which crashed plot2()
Added upload_tab argument to create_interactively(): a full-featured Upload tab supporting CSV, TSV, Excel (XLS/XLSX/ODS/XLSB), SPSS, Stata, SAS, RDS, Parquet, Feather, JSON, YAML, XML, and more, each with format-specific options (delimiter, sheet selection, encoding, NA strings, etc.). Added internal plot2:::install_shiny_deps() for administrators to install all Shiny dependencies at once.
Support y.transform = "reverse"
Support legend.nrow
Support back-to-back (= tornado, = butterfly) plots
Support spider (= radar) plots with type = "spider", using the new coordinate system coord_spider()
Support beeswarm plots with type = "beeswarm" (requires the ggbeeswarm package)
