Generate area-proportional Euler diagrams using numerical optimization. An Euler diagram is a generalization of a Venn diagram, relaxing the criterion that all interactions need to be represented. Diagrams may be fit with ellipses and circles via a wide range of inputs and can be visualized in numerous ways.

**eulerr** generates area-proportional Euler diagrams that display set relationships (intersections, unions, and disjoints) with circles or ellipses. Euler diagrams are Venn diagrams without the requirement that all set interactions be present (whether they are empty or not), which means that, depending on input, **eulerr** sometimes produces Venn diagrams and sometimes not.

With three or more sets intersecting, exact Euler diagrams are often impossible. For such cases **eulerr** attempts to provide a good approximation by numerically tuning the parameters of the ellipses or circles to minimize the error in the resulting diagram. Residuals and goodness of fit statistics are provided to assess whether the resulting diagram can be trusted.

install.packages("eulerr")

devtools::install_github("jolars/eulerr")

eulerr is also available as a shiny app hosted at jolars.co/eulerr.

library(eulerr)# From Wilkinson 2012fit <- euler(c("A" = 4, "B" = 6, "C" = 3, "D" = 2, "E" = 7, "F" = 3,"A&B" = 2, "A&F" = 2, "B&C" = 2, "B&D" = 1,"B&F" = 2, "C&D" = 1, "D&E" = 1, "E&F" = 1,"A&B&F" = 1, "B&C&D" = 1),shape = "ellipse")

We can inspect the goodness-of-fit metrics *diagError* and *stress* for the solution,

fit$stress#> [1] 3.762385e-12fit$diagError#> [1] 3.944839e-07

and plot it

plot(fit)

Please see the introductory vignette for a brief introduction or *eulerr under the hood* for details.

eulerr is open source software, licensed under GPL-3.

eulerr uses semantic versioning.

Please note that this project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms.

eulerr would not be possible without Ben Frederickson's work on venn.js or Leland Wilkinson's venneuler.

`plot.euler()`

has been rewritten completely from scratch, now using a custom**grid**-based implementation rather than**lattice**. As a result, all`panel.*()`

functions and`label()`

have been deprecated as well as arguments`fill_alpha`

,`auto.key`

,`fontface`

,`par.settings`

,`default.prepanel`

,`default.scales`

, and`panel`

. The method for plotting diagrams has also changed---rather than overlaying shapes on top of each other, the diagram is now split into separate polygons using the**polyclip**package. Instead of relying on semi-transparent fills, the colors of the fills are now blended in the CIELab color space (#16).- The default color palette has been resdesigned from scratch to suit the new plot method.
- A new function
`eulerr_options()`

have been provided in order to set default grahical parameters for the diagrams.

- Arguments
`counts`

and`outer_strips`

to`plot.euler()`

are now defunct. `euler()`

now always returns ellipse-based parameters with columns`h`

,`k`

,`a`

,`b`

, and`phi`

, regardless of which shape is used. This item was previously named "coefficients", but it now called "ellipses" instead and a custom`coef.euler()`

method has been added to make cure that`coef()`

still works.- Layouts are now partially normalized so that diagrams will look approximately the same even with different random seeds.

- Providing custom labels to
`quantities`

and`labels`

arguments of`plot.euler()`

now works correctly (#20).

- The last-ditch optimizer switched from
`GenSA::GenSA()`

to`RcppDE::DEoptim()`

. - The optimizer used in all the remaining cases, including all circular
diagrams and initial layouts, was switched back to
`stats::nlm()`

again. - In final optimization, we now use
*stress*instead of residual sums of squares as a target for our optimizer.

`label`

is now a proper generic with an appropriate method (`label.euler()`

).- The
**eulerr under the hood**vignette has received a substantial update.

- Fixed warnings resulting from the deprecated
`counts`

argument in one of the vignettes. - Fixed memcheck errors in the final optimizer.
- Corrected erroneous labeling when
`auto.key = TRUE`

and labels were*not*in alphabetic order. (#15)

- Added the missing %\VignetteEngine{knitr::knitr} to both vignettes. It had mistakenly been left out, which had mangled the resulting vignettes.

- Ellipses are now supported by setting the new argument
`shape = "ellipse"`

in`euler()`

. This functionality accompanies an overhaul of the innards of the function. - Initial optimization function and gradient have been ported to C++.
- The initial optimizer has been switched from
`stats::optim(..., method = "L-BFGS-B")`

to`stats::nlminb()`

. - The final optimizer now falls back to
`GenSA::GenSA()`

when the fit from`nlminb()`

isn't good enough, by default for 3 sets and ellipses, but this behavior can be controlled via a new argument`control`

. - A packing algorithm has been introduced to arrange disjoint clusters of ellipses/circles.
- The label placement algorithm has been rewritten to handle ellipses and been ported to C++. It now uses numerical optimization, which should provide slightly more accurate locations.
- The initial optimizer now uses an analytical Hessian in addition to gradient.

- The initial optimizer now restarts up to 10 times and picks the best fit (unless it is perfect somewhere along the way).
- The default palette has been changed to a fixed palette, still adapted to color deficiency, but with some manual adjustments to, among other things, avoid unnecessary use of color.
- The names of the
`diagError`

and`regionError`

metrics have been changed from`diag_error`

and`region_error`

to reflect the original names. - The coordinates for the centers are now called
*h*and*k*instead of*x*and*y*, respectively. - A new
`label()`

function has been added to extract locations for the overlaps for third party plotting (#10). - The
`counts`

argument to`plot.euler()`

and`panel.euler.labels()`

have been deprecated in favor of the more appropriate`quantities`

. - Argument
`fill_opacity`

in`plot.euler()`

that was deprecated in v2.0.0 has been made defunct.

`eulerr()`

has been replaced with`euler()`

(see update 1.1.0) and made defunct.- There are two new methods for
`euler`

:`euler.list()`

produces diagrams from a list of sample spaces.`euler.table()`

produces diagrams from a`table`

object, as long as there are no dimensions with values greater than 2.

`plot.euler()`

has been rewritten (again) from the ground up to better match other high-level functions from**lattice**. This change is intended to be as smooth as possible and should not make much of a difference to*most*users.- Arguments
`polygon_args`

,`mar`

, and`text_args`

to`plot.euler()`

have been made defunct.

`plot.euler()`

handles conflicting arguments better.- c++ routines in
`eulerr`

now use registration. `euler()`

now allows single sets (#9).- Labels in
`plot.euler()`

now use a bold fontface by default in order to distinguish them from the typeface used for counts. - Argument
`key`

in`plot.euler()`

has been deprecated and replaced with`auto.key`

. Notice that using`key`

does not throw a warning since the argument is used in`lattice::xyplot()`

(which`plot.euler()`

relies on). - Argument
`fill_opacity`

is softly deprecated and has been replaced with`fill_alpha`

for consistency with other lattice functions.

`border`

argument in`plot.euler()`

works again (#7).

`eulerr()`

and its related methods been deprecated and are being replaced by`euler()`

, which takes slightly different input. Notably, the default is now to provide input in the form of disjoint class combinations, rather than unions. This is partly to make the function a drop-in replacement for`venneuler::venneuler`

.`plot.euler()`

has been completely revamped, now interfacing`xyplot()`

from lattice. As a result, arguments`polygon_args`

,`mar`

, and`text_args`

have been deprecated.

- Added a
`counts`

argument to`plot.eulerr`

, which intersections and complements with counts from the original set specificiation (#6). - Added a
`key`

argument to`plot.eulerr`

that prints a legend next to the diagram. - Switched to
`atan2()`

from RcppArmadillo. - Added version requirement for RcppArmadillo.
- Dropped dependency on MASS for computing label placement, replacing it with a faster, geometric algorithm.
- Dropped the cost function argument
`cost`

and now forces the function to use sums of squares, which is more or less equivalent to the cost function from`venneuler`

. - Color palettes in
`plot.euler()`

now chooses colors adapted to color vision deficiency (deuteranopia). With increasingly large numbers of sets, this adaptation is relaxed to make sure that colors are kept visually distinct. `euler()`

now uses`nlm()`

instead of`optim(method = "Nelder-Mead")`

for its final optimization.

- The previous algorithm incorrectly computed loss from unions of sets. It now computes loss from disjoint class combinations.
- Added missing row breaks in
`print.eulerr`

.

- Final optimization routines have been completely rewritten in C++ using Rcpp and RcppArmadillo.
- Switched to the cost function from EulerAPE for the default optimization
target but added the posibility to choose cost function via a
`cost`

argument (currently`eulerAPE`

or`venneuler`

). - Added the option to produce conditional eulerr plots via a
`by`

argument to`eulerr`

. The result is a list of euler diagrams that can be plotted in a grid arrangement via a new plot method. - Improved label placement by using a two-dimensional kernel density estimation instead of means to calculate label centers.

- Cleaned up typos and grammar errors in the
*Introduction to eulerr*vignette. - Added
`mar`

argument to`plot.eulerr`

with a default that produces symmetric margins. - Corrected the implementation of the
`stress`

statistic from venneuler. - Switched to Vogel sampling to generate points to choose label positions from.
- Minor clean up and performance fixes all around.
- Added a
`print.eulerr`

method. - Updated vignette to cover new features and changes.

- The first release.