NEWS.md

In ggdist: Visualizations of Distributions and Uncertainty

ggdist 3.3.2

Major changes:

• The geom_slabinterval() and geom_dotsinterval() families gain "sub-guides", which can be passed to the subguide parameter to create axis annotations for the thickness aesthetic (for slabs) and the dot count (for dots) (#183).
• The weight aesthetic is now supported in stat_slabinterval(), including weighted calculations for densities, CDFs, all interval types (quantile intervals, highest density intervals, and highest density continuous intervals), and all point summaries (mean, median, and mode) (#41). This includes support for the upcoming weighted random variable type in the posterior package.
• Blurry dotplots are now supported using geom_blur_dots(), which accepts an sd aesthetic to set the standard deviation of the blur on each dot. Intervals can also be used in place of blur by passing blur = "interval". This geom is used by the new stat_mcse_dots() to show quantiles along with their error using blur (#63).
• The new breaks_quantiles() histogram breaks function allows the construction of quantile histograms with density_histogram(), stat_histinterval(), etc.
• The color ramp scales (e.g. scale_colour_ramp_continuous(), ...) now use an explicit data type, partial_colour_ramp(), to encode color ramps and their origin colors, and provide the ramp_colours() function for applying colour ramps. This should make it easier to pass explicit color ramps without using scale functions, and for packages building on {ggdist} to use the colour ramp scales (#209).

Minor changes:

• The default histogram bin selection algorithm is now "Scott" instead of "Sturges", as "Sturges" tends to be too conservative (#214).
• The at parameter to stat_spike() (or its names) now determines values of an at computed variable, which can be mapped onto aesthetics via after_stat() to more easily label spikes. (#203; thanks @mattansb for the suggestion).
• The arrow parameter is now supported for intervals in geom_slabinterval() (#206; thanks to @ASKurz for the suggestion).
• The default value of overflow in geom_dotsinterval() is now the new "warn" mode, which works the same as "keep" except that it warns users if the dots will overflow the geometry bounds and suggests solutions (#213).
• Optional arguments to automatically partially-applied functions can now be passed a waiver() to use their default value (see auto_partial()).
• Several dependency reductions: removed {cowplot}, {purrr}, {forcats}, {palmerpenguins}, and {modelr} from Suggests; moved {tidyselect} and {dplyr} from Imports to Suggests. The latter two are only strictly necessary for curve_interval() due to its use of grouped data frames and tidy selection to specify which columns are conditional and which are joint (the use of grouped data frames with point_interval() is less strictly necessary, and not used by stats, so is easier to avoid as an absolute dependency).

Documentation:

• The pkgdown documentation now includes an online article on the thickness aesthetic with comprehensive examples of how slab scaling works (#205).

Bug fixes:

• Ensure Mode() works on analytical constant distributions.
• Various fixes to ensure compatibility with {ggplot2} 3.5.0.

ggdist 3.3.1

New features and enhancements:

• Use derivatives supplied by transformations in scales >= 1.2.2 to make transformations of densities more reliable (r-lib/scales#341).
• New layout = "bar" for geom_dotsinterval() that provides better bar dotplots (with thanks to @sharoz for feedback; #190).
• Bandwidth estimators (including the default, bandwidth_dpi()) now fall back to bandwidth_nrd0() when they fail, with a warning that suggests trying a dotplot or histogram (as these failures tend to happen on data that is not a good candidate for a density plot in the first place) (#196).
• Much faster (C++) implementation of Wilkinson dotplot binning, especially for large dotplots.

Bug fixes:

• Ensure scale_side_mirrored() supports start = "left" and start = "right"
• Ensure geom_spike() draws the point on the correct end of the line depending on side.
• Future-proof guide_rampbar() for ggplot2 > 3.4.2 (#186). Thanks to @teunbrand.
• Future-proof some minor tests for ggplot2 > 3.4.2 (#187).
• Allow the size aesthetic to be overridden for the geom_dots() legend.
• Ensure hdi() supports constants. (#194)

ggdist 3.3.0

Breaking changes: The following changes, mostly due to new default density estimators, may cause some plots on sample data to change. Changes should usually be small, and generally should result in more accurate density estimation. Revert to the old behavior by setting density = density_unbounded(bandwidth = "nrd0").

• stat_slabinterval() now uses density_bounded() as its default density estimator, which uses a bounded density estimator that also estimates the bounds of the data. The default bandwidth estimator is also now bandwidth_dpi(), which is the Sheather-Jones direct plug-in estimator (the same as stats::bw.SJ(..., method = "dpi")). These changes may cause existing charts using densities to change; usually only slightly. These changes should be worth it, as they should drastically improve the accuracy of density estimates, especially on bounded data, and should have little noticeable impact on densities on unbounded data.
• density_bounded() now estimates bounds from the data when not provided (i.e. when one of bounds is NA). See the bounder_ functions (e.g. bounder_cdf(), bounder_cooke()) for more on bounds estimation.
• Improved Mode() and hdi() estimators based on bounded density estimator.

New features and enhancements:

• Improved hdci() estimator using quantile estimation.
• Histograms are now implemented using density_histogram(), a histogram density estimator. Finer-grained control of bin positions is now possible using the breaks argument (including the new breaks_fixed() for manually-specified bin widths) and the align argument (including the new align_boundary() and align_center() for choosing how to align bin positions to reference points). (#118)
• New geom_spike() and stat_spike() for adding spike annotations to slabs created with geom_slabinterval() or stat_slabinterval(). See example in vignette("slabinterval"). (#58, #124)
• parse_dist() now outputs distributional objects in a .dist_obj column in addition to the name-plus-arguments (.dist+.args) format, and these objects respect truncation parameters from prior specifications. This makes it easier to visualize standard deviation priors, for example, giving a better solution to #20.
• marginalize_lkjcorr() adjusts the .dist_obj column output by parse_dist() in addition to the .dist and .args columns.
• geom_lineribbon() now obeys the order aesthetic, allowing you to arbitrarily set the draw order of ribbons (#171). Enabled by this change, stat_lineribbon() now sets order = after_stat(level) by default, making its draw order more correct by ensuring all ribbons of the same level are drawn together.
• Some improved error messages using cli.
• Very experimental adaptive KDE is available through the adapt parameter; note that it is unsupported and both the implementation and interface are highly likely to change.

Deprecations:

• The slab_type parameter for stat_slabinterval() is now deprecated in favor of mapping the corresponding computed variable (pdf or cdf) onto the desired aesthetic. For slab_type = "histogram", use the pdf computed variable combined with the new density_histogram() density estimator (e.g. set density = "histogram"). (#165)

Bug fixes:

• Ensure scale transformations work even when no slab is present; e.g. in stat_interval(). (#168)
• Ensure curve_interval() works with posterior::rvars. (#158)
• geom_lineribbon() draw order is now correct even when some portions of a ribbon has NA widths. (#171)
• Improve the appearance of logical fill conditions at bin edges on histograms. (#175)

ggdist 3.2.1

New features and enhancements:

• Support for non-numeric distributions in stat_slabinterval() and stat_dotsinterval(), including dist_categorical(), dist_bernoulli(), and the upcoming posterior::rvar_factor() type. (#108)
• Various improvements to dotplot layout in geom_dotsinterval():
• new layout = "hex" allows a hexagonal circle-packing style layout (#161).
• new mechanism for smoothing dotplots using the smooth parameter, including smooth = "bounded" / smooth = "unbounded" (for "density dotplots") and smooth = "discrete" / smooth = "bar" (for improved layout of large-n discrete distributions). (#161)
• a better bin/dot-nudging algorithm using constrained optimization (#163)
• new overlaps = "keep" option disables bin/dot nudging in "bin", "hex", and "weave" layouts. This means layout = "weave" with overlaps = "keep" will yield exact dot positions. (#161)
• The "weave" layout now works properly with side = "both"
• fixed binning artifacts when there is high density on the edges, particularly right edges (#144)
• use a max binwidth of 1 for discrete distributions (#159)
• new overflow = "compress" allows layouts to be compressed to fit into the geom bounds if a user-specified binwidth would otherwise cause the dots to exceed the geom bounds. (#162)
• Two new shortcut geoms for geom_dotsinterval(): geom_swarm() and geom_weave(). Both can be used to quickly create "beeswarm"-like plots.
• A new "mirrored" scale for the side aesthetic, scale_side_mirrored(), makes it easier to create mirrored slabs and dotplots. (#142)
• Custom density estimators can now be used with stat_slabinterval() via the density argument, including a new bounded density estimator (density_bounded()). (#113)
• Following the split between size and linewidth aesthetics in ggplot2 3.4, the following aesthetics have been updated (#138):
• interval_size is now linewidth
• slab_size is now slab_linewidth
• in geom_slab(), geom_dots(), and geom_lineribbon(), size is now linewidth
• A new experimental mini domain-specific language for probability expressions in ggdist stats: the Pr_() and p_() functions can be used to generate after_stat() expressions in terms of ggdist computed variables; e.g. aes(thickness = !!Pr_(X <= x)) maps the CDF of the distribution onto the thickness aesthetic; aes(thickness = !!p_(x)) maps the PDF onto the thickness aesthetic. (#160)
• Several function families in ggdist now use "currying" (automatic partial function application). These function families partially apply themselves until all non-optional arguments have been supplied: point_interval(), smooth_..., and density_.... See help("automatic-partial-functions").
• Performance improvements for point_interval() on grouped data frames. (#154)

Documentation:

• Uses of stat() have been replaced with after_stat() to be consistent with the deprecation of stat() in ggplot2 3.4.

ggdist 3.2.0

New features and enhancements:

• Several computed variables in stat_slabinterval() can now be shared across sub-geometries:
• The .width and level computed variables can now be used in slab / dots sub-geometries. These values correspond to the smallest interval computed in the interval sub-geometry containing that portion of the slab. This gives a more flexible alternative to using cut_cdf_qi() to create densities filled according to a set of intervals (this approach which also works on highest-density intervals, which cut_cdf_qi() does not). Examples in vignette("slabinterval") have been updated to use the new approach, and an example has been added to vignette("dotsinterval") showing how to color dots by intervals.
• As an experimental feature (currently a bit fragile) enabled via options(ggdist.experimental.slab_data_in_intervals = TRUE), the pdf and cdf computed variables can now be used in interval sub-geometries to get the PDF and CDF at the point summary. pdf_min, pdf_max, cdf_min, and cdf_max also give the PDF and CDF at the lower and upper ends of the interval. An example in vignette("lineribbon") shows how to use this to make lineribbon gradients whose color approximates density (as opposed to the classic gradient fan chart examples already in that vignette, where color approximates the CDF).
• scale_thickness_shared() is now provided to allow the thickness scale to be shared across geometries, making certain plot types easier to create (e.g. plots of prior and posterior densities together). See vignette("slabinterval") for an example.
• If thickness is less than 0 it is normalized to have a minimum of zero when normalization is turned on; this makes it easier to use slab functions that go below zero. A new example in vignette("slabinterval") shows how to use this to create raindrop plots.
• The stacking order of dots within bins for geom_dotsinterval(layout = "bin") can now be set using the order aesthetic. This makes it possible to create "stacked" dotplots by mapping a discrete variable onto the order aesthetic (#132). As part of this change, bin_dots() now maintains the original data order within bins when layout = "bin". See an example in vignette("dotsinterval").
• A new verbose = TRUE flag in geom_dotsinterval() outputs the selected binwidth in both data units and normalized parent coordinates. This may be useful if you want to start with an automatically-selected bin width and then adjust it manually. Though note: if you just want to scale the selected bin width to fit within a desired area, it is probably better to use scale, and if you want to provide constraints on the bin width, you can pass a 2-vector to binwidth.
• The expand argument in stat_slabinterval() can now take a length-two logical vector to control expansion to the lower and upper limits respectively (#129). Thanks to @teunbrand.
• geom_dotsinterval() now supports the family aesthetic for setting the font used to display its dots (based on a conversation with @gdbassett).
• Experimental guide_rampbar() for creating gradient-like legends for continuous color/fill ramp scales, based on ggplot2::guide_colorbar(). See an example in vignette("lineribbon").

Bug fixes:

• If there are NAs in the thickness aesthetic of a slab, these are now rendered as gaps in the slab (#129).
• Fixed the check for empty x/y scales to avoid extending the scale to cover 0/1 when plotting distributional objects whose bulk lies outside that region (when there is nothing else on the plot).

ggdist 3.1.1

Bug fixes:

• If a string is supplied to the point_interval argument of stat_slabinterval(), a function with that name will be searched for in the calling environment and the ggdist package environment. The latter ensures that stats work when ggdist is loaded but not attached to the search path (#128).

ggdist 3.1.0

New features and enhancements:

• The stat_sample_... and stat_dist_... families of stats have been merged (#83).
• All stat_dist_... stats are deprecated in favor of their stat_... counterparts, which now understand the dist, args, and arg1...arg9 aesthetics.
• xdist and ydist can now be used in place of the dist aesthetic to specify the axis one is mapping a distribution onto (dist may be deprecated in the future).
• Passing dist-like objects to the x or y aesthetics now raise a helpful error message suggesting you probably want to use xdist or ydist.
• Restructured internals for stats and geoms makes it much easier to maintain shortcut geoms and stats, eliminating a large amount of code duplication (#106).
• New expand parameter to stat_slabinterval() allows explicitly setting whether or not the slab is expanded to the limits of the scale (rather than implicitly setting this based on slab_type).
• The point_interval() family of functions can now be passed distributional and posterior::rvar() objects, meaning that means and modes (in addition to medians) and highest-density intervals (in addition to quantile intervals) can now be visualized for analytical distributions.
• As part of this, multivariate distribution objects and rvars will generate a .index column when passed to point_interval() functions (#111). Based on a suggestion from @mitchelloharawild.
• New stat_ribbon() provided as a shortcut stat for stat_lineribbon() with no line (#48). Also, if you supply only an x or y aesthetic to geom_lineribbon(), you will get ribbons without a line (#127).
• One-sided intervals (i.e. quantiles) can now be calculated using ul() (upper limit) or ll() (lower limit), e.g. with point_interval() explicitly or via mean_ll(), median_ll(), mode_ll(), mean_ul(), median_ul(), or mode_ul() (#49).
• Constant distributions are now reliably detected in a variety of situations and rendered as point masses in both density plots and histograms (#103, #32).
• Minor improvements and changes to dotplot layouts that may result in minor changes to the appearance of existing dotplots:
• Minor improvements to automatic bin width selection; the maximum dot stack height should be closer to or equal to scale more often.
• A formerly-internal fudge factor of 1.07 for dot sizes is now exposed as the default value of the dotsize parameter instead of being applied internally. This fudge factor tends (in my opinion) to make dotplots look a bit better due to the visual distance between circles, but is (I think) better used as an explicit value than an implicit one, hence the change. This may create subtle changes to plots that use the dotsize or stackratio parameters, but allows those parameters to have a more precise geometric interpretation.

Documentation:

• New vignette for the stat_dotsinterval() sub-family: vignette("dotsinterval") (#120).
• Vastly improved and expanded documentation for the stat_slabinterval() and geom_slabinterval() family: each shortcut stat/geom now has its own documentation page that comprehensively lists all parameters, aesthetics, and computed variables, including those pulled in via ... from typically-paired geoms. These docs are auto-generated and should be easy to maintain going forward. (#36)
• The stat_lineribbon() and geom_lineribbon() family now also has separate documentation pages with a comprehensive listing of aesthetics and parameters (#107).
• Ridge plot-like example in vignette("slabinterval") using the new expand parameter (#115).

Deprecations and removals:

• The .prob argument, a long-deprecated alias for .width, was removed.
• The limits_function, limits_args, slab_function, slab_args, interval_function, and interval_args arguments to stat_slabinterval() were removed: these were largely internal-use parameters only needed by subclasses of the base class for creating shortcut stats, yet added a lot of noise to the documentation, so these were replaced with the $compute_limits(), $compute_slabs(), and $compute_intervals() methods on the new AbstractStatSlabinterval internal base class.

Bug fixes:

• Improved handling of NAs for analytical distributions.
• Fixed bug where within-bin order of dots in dotplots for "bin" and "weave" layouts could be incorrect with aesthetics mapped at a sub-bin level.
• stackratios that are not equal to 1 are now accounted for in find_dotplot_binwidth() (i.e. automatic dotplot bin width selection).
• Ensure distinct fill colors in lineribbons are still treated as distinct for grouping even if the fill_ramp aesthetic ramps them to the same color.

ggdist 3.0.1

Bug fixes:

• Forward-compatibility fixes for distributional >= 0.2.2.9000 (#91).
• Allow densities for samples of size 1 in stat_sample_slabinterval() (#98).
• Avoid NOTE about missing linearGradient() function on R < 4.1.
• Do not draw legend components for inactive sub-geoms in geom_slabinterval().

ggdist 3.0.0

Breaking changes:

• The positioning of geom_slabinterval() family geoms when using position_dodge() is now slightly different in order to match up with how other geoms are positioned (#85). This may slightly change existing charts that use position = "dodge", and in some cases may cause slabs to be drawn slightly outside plot boundaries, but makes it much easier to combine geom_slabinterval() with other geoms in the expected way. If dodging more similar to the old approach is needed, use the new "justification-preserving dodge", position_dodgejust(), in place of position_dodge().

New features:

• For geom_slabinterval(), side, justification, and scale can now be used as aesthetics instead of parameters, allowing them to vary across slabs within the same geom.
• Varying fills within a slab in geom_slabinterval() can now be drawn as true gradients rather than segmented polygons in R >= 4.1 by setting fill_type = "gradient". This substantially improves the appearance of gradient fills in graphics engines that support it (#44).
• Improved support for discrete distributions:
• stat_dist_slabinterval() and company now detect discrete distributions and display them as histograms (#19).
• geom_dotsinterval() now adjusts bin widths on discrete distributions when they would result in bins that are taller than the allocated space to ensure that they fit within the required space (#42).
• Allow user-specified lower and/or upper bounds on dynamic geom_dotsinterval() bin width by passing a vector of two values to the binwidth parameter.
• The automatic bin selection algorithm used by geom_dotsinterval() has been factored out and exported as find_dotplot_binwidth() and bin_dots() for others to use (#77).
• Previously, curve_interval() used a common (but naive) approach to finding a cutoff on data depth to identify the X% "deepest" curves, simply taking the envelope around the X% quantile of curves ranked by depth. This is quite conservative and tends to create intervals that are too wide; curve_interval() now searches for a cutoff in data depth such that X% of curves are contained within its envelope (#67).
• point_interval() and company now accept distributional objects and posterior::rvar()s (full support for distributional objects requires distributional > 0.2.2).
• Reduce dependencies substantially, making the geoms more suitable for use by other packages (thanks to Brenton Wiernik for the help).

New documentation:

• Substantial improvements to the documentation of aesthetics and computed variables in geom_slabinterval(), stat_slabinterval(), and company, listing all custom aesthetics, computed variables, and their usage.

• Several new examples in vignette("slabinterval"), including "rain cloud" plots and an example of histograms for discrete analytical distributions.

Bug fixes:

• Ensure stat_dist_slabinterval() preserves group order (#88).
• Improve test coverage up to ~96%.
• Restore computed variable n for stat_sample_slabinterval().
• Various improvements in correct NA handling across the geoms (#74, #51).

ggdist 2.4.1

New features:

• Added "weave" and "swarm" layouts for dots geoms (#64). These provide alternative layouts that keep datapoints in their actual positions on the data axis. The "weave" layout maintains rows but not columns and works well for quantile dotplots; the "swarm" layout uses the "compactswarm" method from beeswarm::beeswarm() (courtesy James Trimble) and works well on sample data. See the dotplot section of vignette("slabinterval") for comparisons.
• Allow the use of unit() to specify bin widths manually for dots geoms and stats, which can be helpful when you need dotplots across facets to have the same bin width (#53).

New documentation:

• Add example of lineribbon gradients using fill_ramp in vignette("lineribbon").
• Add example of Tukey-like pencils in vignette("slabinterval").
• Add example of two slab used together (densities and dotplots to make "rain clouds") in vignette("slabinterval").

Bug fixes:

• Fix issues with ggplot2 3.3.4 (#72) and vdiffr 1.0.
• Handle interactions between alpha and fill/color properly when not set by user (#62).
• Use step function for all ECDFs, which should also fix constant CDFs (#55).
• Move fda to suggests as it brings in a large number of dependencies and is rarely used.
• Use trimmed density for mode estimation (#57).

ggdist 2.4.0

New features:

• add pdf and cdf computed variables for the stat_sample_slabinterval() subfamily. See new examples of usage in the last section of vignette("slabinterval"). (#11)
• add cut_cdf_qi() for creating (amongst other things) interval-filled halfeyes, in the style of bayesplot::mcmc_areas() (#11)
• add fill_ramp and color_ramp scales to geom_slabinterval() and geom_lineribbon() families, making it easier to separate group colors from interval/density/CDF colors. See new examples in vignette("slabinterval"), vignette("lineribbon"), and vignette("freq-uncertainty-vis"). (#16)
• auto-detect finite limits on analytical distributions, so (e.g.) distributions like the beta will not cut off tails close to 0 or 1. (#18)
• add brms::brmsprior implementation for parse_dist() (#34)

New documentation:

• vignette("freq-uncertainty-vis") now uses distributional::dist_student_t() (#14)
• add references to fuzzygrams in vignette("slabinterval") (#23)
• add examples with separate positioning of slab and interval (#27)
• add discussion of limitations of curvewise intervals to vignette("lineribbon") (#22)
• soft-deprecate usage of interval_size_range argument in docs (#35)
• initial versions of some cheat sheets

Bug fixes:

• add limited na.rm support to curve_interval() (#22)
• use analytical instead of numerical derivatives on scale transformations where possible, improving reliability.

ggdist 2.3.0

New features and documentation:

• Add curve_interval() for generating curvewise (joint) intervals for curve boxplots (#22)
• Add vignette("lineribbon") describing geom_lineribbon(), stat_lineribbon(), stat_dist_lineribbon(), and curve_interval().

Bug fixes:

• Support dist aesthetics that are factors (#25)
• Fix slab drawing order for overlapping (ggridges-style) slabs (#30)
• Workaround for changes to {distributional} distribution functions until #31 is fixed.

ggdist 2.2.0

• Support for distributional, including new examples in vignette("slabinterval") (#14).
• stat_dist_... geoms now calculate pdf and cdf columns to allow mashup geoms that involve both functions, such as Correll-style gradient plots combined with violins, as in Helske et al. (#11).
• stat_dist_... geoms should now work with gganimate (#15).
• Examples updated to fix errors introduced by broom::augment() defaulting to se_fit = FALSE.

ggdist 2.1.1

• Initial split from tidybayes: ggdist now contains all stats/geoms from tidybayes (except deprecated ones), all support functions for stats/geoms (such as point_interval()), vignette("slabinterval"), and vignette("freq-uncertainty-vis"). Tidybayes will retain all other functions, and will re-export all ggdist functions for now.
• All stats and geoms now support automatic orientation determination. Thus, all h-suffix geoms are now deprecated. Those geoms have been left in tidybayes and give a deprecation warning when used; they cannot be used from ggdist directly.
• geom_interval(), geom_pointinterval(), and geom_lineribbon() no longer automatically set the ymin and ymax aesthetics if .lower or .upper are present in the data. This allows them to work better with automatic orientation detection (and was a bad feature to have existed in the first place anyway). The deprecated tidybayes::geom_intervalh() and tidybayes::geom_pointintervalh() still automatically set those aesthetics, since they are deprecated anyway (so supporting the old behavior is fine in these functions).
• geom_lineribbon()/stat_lineribbon() now supports a step argument for creating stepped lineribbons. H/T to Solomon Kurz for the suggestion.
• ggdist now has its own implementation of the scaled and shifted Student's t distribution (dstudent_t(), qstudent_t(), etc), since it is very useful for visualizing confidence distributions.

ggdist documentation built on July 4, 2024, 9:08 a.m.