[R-pkgs] ggplot2 0.9.0
hadley at rice.edu
Fri Mar 2 14:04:16 CET 2012
ggplot2 is a plotting system for R, based on the grammar of graphics,
which tries to take the good parts of base and lattice graphics and
avoid bad parts. It takes care of many of the fiddly details
that make plotting a hassle (like drawing legends) as well as
providing a powerful model of graphics that makes it easy to produce
complex multi-layered graphics.
Find out more at http://had.co.nz/ggplot2, and check out the nearly 500
examples of ggplot in use. If you're interested, you can also sign up to
the ggplot2 mailing list at http://groups.google.com/group/ggplot2, or track
development at http://github.com/hadley/ggplot2
ggplot2 0.9.0 has an extensive set of changes, summarised below and
described in detail in a 40 page transition guide:
* `annotation_custom`: a new geom intended for use as static annnotations that
are the same in every panel. Can be used to add inset plots, tables, and
other grid-based decorations inside the plot area (Contributed by Baptiste
* `geom_map`: a new special case of `geom_polygon` useful when you are drawing
maps, particularly choropleth maps. It is matched with `annotation_map`, an
even faster special case when you want the same map drawn in each panel.
* `geom_raster` is a special case of `geom_tile` for equally sized rectangular
tiles. It uses the raster functionality of R graphics devices for massively
increased speed and much decreased file sizes. It is matched with
`annotation_raster`, an even faster special case, for when you want to draw
the same raster in each panel.
* `geom_violin`: an implementation of violin plots, which are a way of
visualizing kernel density estimates. (Thanks to Winston Chang)
* `geom_dotplot`: dot plots, as described in Wilkinson (1999). To bin the
data, it uses `stat_bindot` to bin the data, which has two methods: histodot
and dot-density. Histodot binning uses fixed-width bins just like
`stat_bin`, while dot-density binning uses variable-width bins. A new grob,
`grob_dotstack` is used to render the dots. (Thanks to Winston Chang)
* New fortify methods have been added for objects produced by the `multcomp`
* `stat_summary2d` and `stat_summary_hex`. These are work like `stat_bin2d`
and stat_binhex but allow any summarisation function (instead of just
count). They are 2d analogs of `stat_summary`
* `facet_grid`: The space argument now supports `free_x` and `free_y` next to
`free` and `fixed, this allows the user to adjust the spatial scaling of the
facets in either the x or y direction. This is especially useful when the
scales are very different. In this case space = `free` could make some
facets very small. (Thanks to Willem Ligtenberg)
* Thorough clean up and checking, including documenting all arguments, adding
systematic cross-references, and adding commonly requested examples. Thanks
to Jake Russ and Dennis Murphy for the help.
* Complete series of aesthetics pages (grouped subsets of aesthetics) with
examples of how to use the major ones, see e.g. `?fill`, `?shape`, `?x`,
* Added a complete list of theme opts with usage examples in `?opts`
* Added "translate" pages to demonstrate usage between qplot and ggplot, GPL,
base and lattice graphics: `?translate_qplot_base`, `?translate_qplot_gpl`,
* Scales have been rewritten to use the new `scales` package, which does a
much better job at defining what a scale is and making it easier for you to
create your own scales. Scales should now behave much more consistently, and
it should be easier for me to add new features in the future.
* `breaks` parameter can now be a function, which will be passed the scale
limits and expected to return a character vector of breaks
* `labels` parameter can now be a function - this replaces the previous
formatter function that only some scales possessed, and the `major` argument
to the data time scales. This function should take a vector of breaks as
input, and return a character vector or list of expressions as output. See
`comma_format`, `dollar_format`, `percent_format`, `scientific_format`,
`parse_format` and `math_format` for examples
* Transformations are now provided by the scales package - see `?trans_new`
for list of available transformations, and how to create your own. The
transformations in this package should do a better job at computing default
* Transformations for continuous scales are now detected automatically when
the default scales are added. This ensures that dates and date times will
display correctly when used for any aesthetic - previously they only worked
with position scales. The system is now also easier to extend to new types
of continuous data that you might want to plot. (Fixes #48)
* All scales now accept a `na.value` parameter which provides an aesthetic
value to be used for `NA` values in the data. Colour/fill scales default to
grey, which should stand out as different from non-missing values.
* The new `oob` (out of bounds) parameter controls how scales deals with
values outside the limits. The default action is `censor` - see `clip` for
* Only `scale_x_log10`, `scale_x_sqrt` and `scale_x_reverse` provided as
convenience functions for x and y scales. Use e.g. `scale_x_continuous(trans
= "log")` to access others
* `set_default_scale` has been removed. If you want to change the default
scale for an aesthetic, just create a function called
`scale_aesthetic_continuous` or `scale_aesthetic_discrete` that returns the
scale that you want. For example:
p <- qplot(mpg, wt, data = mtcars, colour = factor(cyl))
scale_colour_discrete <- scale_colour_brewer
* Scales now automatically shrink to what is actually displayed on the plot,
not the underlying data used for statistical transformation. If you want the
old behaviour, supply `shrink = FALSE` to the facetting specification.
* `scale_colour_gradient` and `scale_fill_gradient` now use a colour scheme
with constant hue but varying chroma and luminance. This is better because
it creates a natural ordering inline with the order of the colour values.
* Converted from proto to S3 objects, and class methods (somewhat) documented
in `facet.r`. This should make it easier to develop new types of facetting
* The new `facet_null` specification is applied in the default case of no
faceting. This special case is implemented more efficiently and results in
substantial performance improvements for non-facetted plots.
* Facetting variables will no longer interfere with aesthetic mappings -
`facet_wrap(~ colour)` will no longer affect the colour of points.
* ggplot2 has moved away from the two (!!) homegrown documentation systems
that it previously relied on, and now uses roxygen extensively. The current
downside is that this means that ggplot2 website can no longer be updated,
but I hope work with the `helpr` package will resolve that shortly.
* ggplot2 now uses a `NAMESPACE`, and only exports functions that should be
user visible - this should make it play considerably more nicely with other
packages in the R ecosystem. Note that this means you now need to explicitly
load `plyr` (and other packages) if you are using them elsewhere in your
* ggplot2 now has a start on a set of automated tests. As this test suite
expands it will help me ensure that bugs stay fixed, and that old bugs don't
come back in new versions. A test suite also gives me more confidence when
I'm modifying code, which should help with general code quality.
* Converted from proto to S3 objects, and class methods (somewhat) documented
in `coord.r`. This should make it easier to develop new types of coordinate
* Added a new method `coord_range` for finding the x and y range even after
coordinates have been transformed to other names (eg., theta and r). (Thanks
to Winston Chang)
* When printing a ggplot2 object, the rendered plot information is returned
invisibly. You can capture this with (e.g.) `x <- print(qplot(mpg, wt, data
= mtcars))` and in the future will be able to use it to get information
about the plot computations, such as the range of all the scales, and the
exact data that is plotted.
* Drawing a plot takes place in three documented steps: `ggplot_build` which
creates a list of data frames ready for rendering builds, `ggplot_gtable`
which creates a `gtable` of grobs, and `grid.draw` which renders the grobs
on screen. Each of these returns a data structure which should be useful for
understanding and modifying the rendered plot. This is still a work in
progress, so please ask questions if anything is confusing.
* The `drop` and `keep` parameters to `ggsave` and `print.ggplot` have been
dropped, as the data structure returned by `ggplot_gtable` is sufficiently
rich enough to remove the need for them.
* Axis labels are now centred underneath the panels (not the whole plot), and
stick close to the panels regardless of the aspect ratio.
* Guides (particularly legends) have been rewritten by Kohske Takahashi to
provide considerably more layout flexibility.
* `guide_legend` now supports multi-row/column legend and reversed order,
gives more flexible positioning of title and label, and can override
aesthetics settings. This is useful, for example, when alpha value in a
panel is very low but you want to show vivid legend.
* `guide_colorbar` is a guide specially for continuous colour scales as
produced by colour and fill scales.
* `geom_text` now supports `fontfamily`, `fontface`, and `lineheight`
aesthetics for finer control over text display. (Thanks to Kohske Takahashi
for the patch. Fixes #60)
* `collide`, which powers `position_dodge` and `position_stack`, now does not
error on single x values (Thanks to Brian Diggs for a fix. #157)
* `...` in `ggplot` now passed on to `fortify` method when used with an object
other than a data frame
* `geom_boxplot`: outlier colour and shape now default to values set by the
aesthetic mapping (thanks to suggestion by Ben Bolker), the width of the
median line is now `fatten` times the width of the other lines (thanks to
suggestion by Di Cook), and the line type can now be set. Notched box
plots are now supported by setting `notch = TRUE` (thanks to Winston Chang
for the patch).
* `ggsave` can work with cm and mm `units` (Thanks to patch from Jean-Olivier
* `scale_shape` finally returns an error when you try and use it with a
* `stat_contour` no longer errors if all breaks outside z range (fixes #195).
* `geom_text` remove rows with missing values with warning (fixes #191)
* New generic function `autoplot` for the creation of complete plots
specific to a given data structure. Default implementation throws
an error. It is designed to have implementations provided by other
packages. (Thanks to suggestion by Brian Diggs)
* `ggpcp` loses the `scale` argument because it relied on reshape(1) code
* `map_data` passes `...` on to `maps::map` (Fixes #223)
* `coord_fixed` accepts `xlim` and `ylim` parameters to zoom in on x and y
scales (Fixes #91)
* ggplot2 will occasionally display a useful hint or tip on startup. Use
`suppressPackageStartupMessages` to eliminate
* `stat_binhex` uses correct bin width for computing y axis bounds. (Fixes
#299, thanks to Dave Henderson for bug report and fix.)
* `stat_smooth` now adjusts confidence intervals from `loess` using a
* `stat_smooth` reports what method is used when method is "auto". It also
picks the method based on the size of the largest group, not individually by
group. (Thanks to Winston Chang)
* `stat_bin` and `geom_histogram` now use right-open, left-closed intervals by
default. Use `right = TRUE` to return to previous behaviour.
* `geom_vline`, `geom_hline`, and `geom_abline` now work with non-Cartesian
coordinate systems. (Thanks to Winston Chang)
Assistant Professor / Dobelman Family Junior Chair
Department of Statistics / Rice University
More information about the R-packages