Make the bayesplot default theme optional?

[gavinsimpson]

At the moment, from what I can tell having only had a brief look at the code, you are hard coding thebayesplot theme_default() ggplot theme when each plot is created. This makes it tedious to override by the user (they must add their own + theme_foo() layers to every plot they create). Further, from a previous Issue I understand this might not work in all cases in bayesplot as some plots are actually multiple plots arranged on the device.

Instead, you can use the theme_set() functionality of ggplot2 to set the default theme when the package is attached. The cowplot package, for example, uses this to set its default theme. This requires an .onAttach() function to be defined in the package R source somewhere with the following:

.onAttach <- function(libname, pkgname) {
  ggplot2::theme_set(bayesplot::theme_default())
}

All instances of + theme_default() would also need to be removed.

Doing this won't change the output from plots unless a user decides to change the theme used with by their own explicit call to theme_set() after bayesplot is loaded.

[jgabry]

Hi Gavin, thanks for the suggestion! Will this proposed method result in changing the plotting theme for any ggplot2 plot created by the user regardless of whether or not it is from bayesplot? That was my original motivation for not doing it this way, i.e. to try to keep bayesplot stuff only affecting bayesplot stuff. If I'm wrong about that or if I'm right but users don't mind that behavior then I'd be open to switching to what you propose.

…

On Sun, Apr 23, 2017 at 12:52 PM Gavin Simpson ***@***.***> wrote: At the moment, from what I can tell having only had a brief look at the code, you are hard coding the*bayesplot* theme_default() ggplot theme when each plot is created. This makes it tedious to override by the user (they must add their own + theme_foo() layers to *every* plot they create). Further, from a previous Issue I understand this might not work in all cases in *bayesplot* as some plots are actually multiple plots arranged on the device. Instead, you can use the theme_set() functionality of *ggplot2* to set the default theme when the package is attached. The *cowplot* package, for example, uses this to set its default theme. This requires an .onAttach() function to be defined in the package R source somewhere with the following: .onAttach <- function(libname, pkgname) { ggplot2::theme_set(bayesplot::theme_default()) } All instances of + theme_default() would also need to be removed. Doing this won't change the output from plots *unless* a user decides to change the theme used with by their own explicit call to theme_set() *after* *bayesplot* is loaded. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#87>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHb4Q70_j22znQrrYdtNMInf7ogGPgrkks5ry4HggaJpZM4NFdyE> .

[gavinsimpson]

@jgabry Yes, you are right; setting the theme globally as I suggested will affect every plot made with ggplot unless the plot code explicitly overrides the theme. However, that's easily and trivially fixed by having the user call theme_set().

If people have the pkg loaded (i.e. it's attached not just being imported from) it's probably because they want to use the package, so the number of people affected by my suggested change is likely low.

On the other hand the solution used in bayesplot isn't easy to work around; a user has to add + theme() etc to each plot, if they can affect the theme at all.

If you were to flag the issue, as cowplot does in documentation and in the README etc, I doubt this change would cause too much grief, but it would make modifying the look and feel of the plots trivial.

[jgabry]

Ok I think I'm persuaded. Shouldn't be hard to change so I'll try to get it in for the next release. Thanks for following up.

…

On Mon, Apr 24, 2017 at 2:21 PM Gavin Simpson ***@***.***> wrote: @jgabry <https://github.com/jgabry> Yes, you are right; setting the theme globally as I suggested will affect every plot made with ggplot unless the plot code explicitly overrides the theme. However, that's easily and trivially fixed by having the user call theme_set(). If people have the pkg loaded (i.e. it's attached not just being imported from) it's probably because they want to use the package, so the number of people affected by my suggested change is likely low. On the other hand the solution used in *bayesplot* isn't easy to work around; a user has to add + theme() etc to each plot, if they can affect the theme at all. If you were to flag the issue, as *cowplot* does in documentation and in the README etc, I doubt this change would cause too much grief, but it would make modifying the look and feel of the plots trivial. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#87 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHb4Q5SM5QElqLOVeBsww1RhqiYpMMV8ks5rzOgWgaJpZM4NFdyE> .

[jgabry]

PR #95 merged so closing this

[gavinsimpson]

Thanks for implementing this wish @jgabry!

[jgabry]

Thanks for suggesting it!

…

On Tue, Jul 25, 2017 at 6:15 PM, Gavin Simpson ***@***.***> wrote: Thanks for implementing this wish @jgabry <https://github.com/jgabry>! — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#87 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHb4Q3P9_LGFSA_qQP8EwjmnBGX_G1-Zks5sRmj1gaJpZM4NFdyE> .