An implementation of an interactive grammar of graphics, taking the best parts of 'ggplot2', combining them with the reactive framework of 'shiny' and drawing web graphics using 'vega'.
The goal of ggvis is to make it easy to describe interactive web graphics in R. It combines:
a grammar of graphics from ggplot2,
reactive programming from shiny, and
data transformation pipelines from dplyr.
ggvis graphics are rendered with vega, so you can generate both raster graphics with HTML5 canvas and vector graphics with svg. ggvis is less flexible than raw d3 or vega, but is much more succinct and is tailored to the needs of exploratory data analysis.
If you find a bug, please file a minimal reproducible example at http://github.com/rstudio/ggvis/issues. If you're not sure if something is a bug, you'd like to discuss new features or have any other questions about ggvis, please join us on the mailing list: https://groups.google.com/group/ggvis.
Install the latest release version from CRAN with:
Install the latest development version with:
# install.packages("devtools")devtools::install_github("hadley/lazyeval", build_vignettes = FALSE)devtools::install_github("hadley/dplyr", build_vignettes = FALSE)devtools::install_github("rstudio/ggvis", build_vignettes = FALSE)
You construct a visualisation by piping pieces together with
%>%. The pipeline starts with a data set, flows into
ggvis() to specify default visual properties, then layers on some visual elements:
mtcars %>% ggvis(~mpg, ~wt) %>% layer_points()
The vignettes, available from http://ggvis.rstudio.com/, provide many more details. Start with the introduction, then work your way through the more advanced topics. Also check out the
various demos in the
demo/ directory. See the basics in
then check out the the coolest demos,
Switched from stripped-down build of jQuery UI to a full build. (#410)
Fixed problems for R CMD check in R 3.3.0.
Remove vignettes due to R CMD check hanging.
ggvis plots can now resize their width to the containing div, with
set_options(width = "auto"). Height can be set automatically as well, but
it will only work properly if the containing div has a fixed height, due to
the way that web browsers do vertical layout. (#316, #374, #381)
compatible with dplyr 0.4.2
input_slider has been updated to work with Shiny 0.11.
The parse spec and update events now happen in the correct order. This fixed an issue with plots flashing. (#351)
Pointer events are now allowed in tooltips (#349)
Updated to Vega 1.4.3 and D3 3.5.2.
Startup messages are now shown only one in ten times. (#302)
Added new dplyr verbs:
ggvis now gives a warning when key prop values are not unique. (#295)
Boxplots are now supported, with
Much better support for data objects with zero rows.
Added support for displaying ggvis plots in dynamic UI in Shiny apps. (#165)
width instead of
compute_bin() now defaults to
pad = FALSE
compute_model_predictions() always returns a result, even if there's an
filter() is no longer imported and re-exported from dplyr. This
means that to use
filter() with ggvis object you'll need to
make sure to load dplyr first.
compute_smooth() supports more complex formulas. (#209)
compute_count() now preserve date and time properties.
export_svg() now work. This requires node.js, and vega
must be installed via npm.
Legend hiding is fixed. (#218)
count_vector() preserves the order of factor levels. (#223)
compute_bin() now ignores NA's. (#148)
layer_bars() now uses correctly uses
fill prop when it is passed to the
function, and not inherited. (#201)
compute_count() drops unused factor levels. (#201)
compute_stack() no longer give warnings and errors for
zero-row data frames. (#211)
Range calculation for zero-length vectors now returns NULL instead of throwing an error.
Objects imported from the magritter and dplyr packages are now properly re-exported.
Using "." in column names now works. (#246)
:=, to avoid possible conflict with data.table.
Updated to Vega 1.4.2. (#193 and #217)
Switched from RJSONIO to jsonlite.
Switched to the new non-standard argument evaluation strategy from dplyr 0.3, using the new lazyeval package.
add_guide_legend() have been replaced by
add_legend(). Also, the interface for
When marks with a
band() prop are added, the appropriate scale is
automatically set to have
points = FALSE. (#128)
Continuous scales have a multiplicative expansion factor added by default,
expand parameter of scale functions.
Relative x and y scales for positioning of graphical elements can be added
Added support for
Added support for controlling width and height of image marks.
prop() objects have been modified so that they always record which scale
qvis(): now the default behaviour of
ggvis() is to add
layer_guess() if there are no layers on the plot already.
add_dscale() has been replaced with
scale_ordinal(), and similar.
Reactive expressions can be used for scale domains. This allows the scale domain to change dynamically.
Axis and legend properties are fixed. (#90)
Histograms allow stacking.
Dynamic plots now with with by_group. (#71)
Gear icon displays properly in Windows. (#159)
layer_bars() are now symmetrical about the x tick positions.
singular() and corresponding
scale_singular() make it easier to
draw plots where x or y are constant (and hence uninteresting), such as
for a 1d dot plot (#127).
pad argument to control whether empty bins
on either side of the data extents are added. This is useful for frequency
polygons and to ensure that histograms don't jam up against the axes.
The main change is that ggvis now uses a functional approach to building plots. Instead of doing:
ggvis(mtcars, props(~wt, ~mpg)) + layer_point()
You now do:
layer_points(ggvis(mtcars, ~wt, ~mpg))
This is a bit clunky, but we streamline it by using the pipe operator (
%>%, from magrittr):
mtcars %>% ggvis(~wt, ~mpg) %>% layer_points()
We think that this change will make it a little easier to create plots, and just as importantly, it's made the internals of ggvis much much simpler (so now we actually understand how it works!). As part of these changes:
We now have a better idea of how layers should work. These are the "magic"
bits of ggvis - they can inspect the current state of the plot, the data and
the visual properties and decide what to do. For an example, take a look at
layer_guess() which implements the most important parts of
guessing which type of layer to use to display the data.
ggvis() and all layer functions now take props directly - you no longer
need to use
props() in everyday work.
You can seamlessly use data transformations from dplyr: that means that
group_by() to define grouping in the plot, and you can use
arrange() both inside and
outside of visualisations. See
ggvis?dplyr for more examples.
Data transformations are now handled by
compute_*() functions. These
are S3 generics with methods for data frames, grouped data frames and
ggvis objects. This means that any transformation done by ggvis for
a visualisation (e.g. smoothing) can also be done on ordinary datasets so
you can see exactly what variables are being created.
It is possible to extract all the data objects, including those that are
created by a transformation function, with the
get_data() function. This
makes it easier to inspect and understand what's happening to your data.
explain() function shows the structure of the ggvis object in a
handle_brush() allow you to connect callbacks to important ggvis events.
A fully reactive interface will follow in the future.
The process of embedding ggvis plots in shiny apps has been overhauled and
simplified. See details in
ggvis?shiny and sample apples in
A new built-in dataset: cocaine, recording cocaine seizures in the US in 2007. We plan to transition our dummy examples that use mtcars to something more useful/informative over time.