[R-sig-teaching] Choice of graphics package
Randall Pruim
rpruim at calvin.edu
Sat Nov 28 22:43:26 CET 2015
Thanks again.
I’m glad to learn that this is available, even if barely documented. For my main use case (wanting to control colors, etc. without cluttering up code I show students), this solution is sufficient.
One further refinement of your example:
scale_fill_discrete <- function(values = c("orange", "purple","pink", "cyan”), …)
scale_fill_manual(values = values, …)
That way, if scale_fill_discrete() happens to be called with the values argument set, it will do what is (probably) the right thing. But since I don’t know exactly where all this gets called, and with which arguments, I don’t know whether this matters or not.
—rjp
On Nov 28, 2015, at 1:35 PM, Ista Zahn <istazahn at gmail.com<mailto:istazahn at gmail.com>> wrote:
On Nov 28, 2015 12:56 PM, "Randall Pruim" <rpruim at calvin.edu<mailto:rpruim at calvin.edu>> wrote:
>
> Thanks, Ista.
>
> Is this documented anywhere?
Sort of -- see https://github.com/hadley/ggplot2/blob/e9532537c9eb4f35af41073ed4aba8eca6ba2133/NEWS#L536 and https://groups.google.com/forum/m/#!topic/ggplot2-dev/_IOAWcKB6DU
>
> I took a look on github and see that scale_fill_discrete is defined to be
>
> scale_fill_discrete <- scale_fill_hue
>
> in zxx.R, and your solution is masking the function defined there.
I think the idea is that scale_fill_discrete should be defined as whatever you want the default fill function to be.
>
> If you type
>
> ?scale_fill_discrete
>
> you land at a man page that does not even mention scale_fill_discrete.
Yeah, ggplot2 documentation leaves something to be desired in places. I think documentation improvements are slated for the next release.
>
> Your updated example works, but I’m wondering how officially it is supported. Also, it seems dangerous to mask a function with one that completely ignores its inputs.
Indeed, we should do
scale_fill_discrete <- function(...) scale_fill_manual(values =
c("orange", "purple","pink", "cyan"), ...)
I’d need to do more code snooping (or find the relevant documentation) to know whether this will break something somewhere else.
I don't see why it would (Hadley recommended it after all) , though I'm ready to concede that this isn't as easy as it should be.
Best,
Ista
>
> —rjp
>
>
>> On Nov 28, 2015, at 9:27 AM, Ista Zahn <istazahn at gmail.com<mailto:istazahn at gmail.com>> wrote:
>>
>> Hi rjp,
>>
>> Arg, I totally botched the example last night. Here is a working
>> example; I think you're right that this is easier in lattice, but it
>> is definitely not true that you need to add this stuff to each plot.
>>
>> library(ggplot2)
>>
>> theme_set(theme_bw() +
>> theme(rect = element_rect(fill = "gray10"),
>> panel.background = element_rect(fill = "gray10")))
>>
>> scale_fill_discrete <- function(...) scale_fill_manual(values =
>> c("orange", "purple",
>> "pink", "cyan"))
>>
>> ggplot(mtcars, aes(x = hp, fill = factor(gear))) + geom_bar()
>>
>> Best,
>> Ista
>>
>>
>> On Sat, Nov 28, 2015 at 9:03 AM, Randall Pruim <rpruim at calvin.edu<mailto:rpruim at calvin.edu>> wrote:
>>>
>>> When I run your code, the colors in my plot are not set to the colors chosen
>>> unless I do
>>>
>>> ggplot(mtcars, aes(x = hp, fill = factor(gear))) + geom_bar() +
>>> scale_discrete
>>>
>>> Did your code example really work for you?
>>>
>>> I think the point is that lattice and ggplot2 differ in terms of what is
>>> called part of the theme. In lattice, the colors used for discrete
>>> variables are part of the theme, and so can be included in the global
>>> default. In ggplot2, they are not part of the theme and cannot be set with
>>> theme_set().
>>>
>>> I get that you can save your scales information as an object and add it on
>>> to any plot so you only have to specify the colors (or whatever) once. But
>>> I think you still need to add the scale information onto each plot.
>>>
>>> —rjp
>>>
>>>
>>> On Nov 27, 2015, at 9:18 PM, Ista Zahn <istazahn at gmail.com<mailto:istazahn at gmail.com>> wrote:
>>>
>>> Woops, messed up the example! I meant
>>>
>>> theme_set(theme_bw() +
>>> theme(rect = element_rect(fill = "gray10")))
>>>
>>> scale_discrete <- scale_fill_manual(values = c("orange", "purple",
>>> "pink", "cyan"))
>>>
>>> ggplot(mtcars, aes(x = hp, fill = factor(gear))) + geom_bar()
>>>
>>> i.e., the whole point is you don't need to add this stuff to every plot.
>>>
>>> Best,
>>> Ista
>>>
>>> On Fri, Nov 27, 2015 at 9:13 PM, Ista Zahn <istazahn at gmail.com<mailto:istazahn at gmail.com>> wrote:
>>>
>>> Hi Randal,
>>>
>>> To be honest I don't know the lattice theme system at all, so it is
>>> entirely possible that you can easily do things there that are more
>>> complicated in ggplot2. However, it is easy to set theme and scale[1]
>>> defaults in ggplot2, e.g.,
>>>
>>> library(ggplot2)
>>>
>>> theme_set(theme_bw() +
>>> theme(rect = element_rect(fill = "gray10")))
>>>
>>> scale_discrete <- scale_fill_manual(values = c("orange", "purple",
>>> "pink", "cyan"))
>>>
>>> ggplot(mtcars, aes(x = hp, fill = factor(gear))) + geom_bar() +
>>> theme_bw() +
>>> scale_fill_manual(values = c("orange", "purple", "pink", "cyan"))
>>>
>>>
>>> Best,
>>> Ista
>>>
>>> [1] In ggplot2 theme elements are distinct from scales, and AFAIK they
>>> have to be set separately.
>>>
>>> On Fri, Nov 27, 2015 at 6:24 PM, Randall Pruim <rpruim at calvin.edu<mailto:rpruim at calvin.edu>> wrote:
>>>
>>>
>>> On Nov 27, 2015, at 3:37 PM, Ista Zahn <istazahn at gmail.com<mailto:istazahn at gmail.com>> wrote:
>>>
>>> (Theming in ggplot2 generally requires you to do something to each plot,
>>> one of the downsides of ggplot2.)
>>>
>>>
>>> That is completely untrue. Whatever downsides ggplot2 may have, this is not
>>> one of them.
>>>
>>>
>>> Ista,
>>>
>>> My guess is that you didn’t understand my claim. Either that or you know
>>> something about ggplot2 that I’ve not been able to locate anywhere. (Since
>>> Hadley is on this list, he can chime in if I’m missing something.)
>>>
>>> So let me explain what I mean with an example.
>>>
>>> Suppose I want to create six plots of various sorts. Each of them uses
>>> color according to a factor a with three levels. That's groups = a in
>>> lattice and colour = a in ggplot2. So far so good.
>>>
>>> But after I have created all of these plots with the default color choices,
>>> I decide I want to have the colors be blue, red, and 50% gray. In lattice,
>>> I can put this information into the default theme with
>>>
>>> trellis.par.set( superpose.symbol = list(col = c("blue", "red", "gray50")) )
>>>
>>> and without any adjustments to any of the plotting code, all the plots will
>>> update to the new color scheme. show.settings() will even show me what my
>>> default theme looks like.
>>>
>>> One line of code changes in one place and all the plots are using the new
>>> color scheme. And if my new choices are choices I will use frequently, I
>>> can put the theme into a package and do something like
>>>
>>> trellis.par.set(theme = theme.mosaic())
>>>
>>> There is also the option to add these themes to individual plots using the
>>> par.settings argument — this is more like the ggplot2 way.
>>>
>>> In ggplot2, I need to add on scale_colour_manual() to EACH PLOT, or write
>>> some sort of wrapper that does the job for me. In either case, if I started
>>> out using the defaults and want to make document-wide changes, I need to
>>> edit EACH plot to get the desired affect. In lattice, this is one line of
>>> code and all the plots are good to go.
>>>
>>> Last I looked for this, ggplot2 did not provide a way to set these sorts of
>>> defaults and generally prefers a system where all of this sort of theming is
>>> located local to each plot. So plots end up having a bunch of theme stuff
>>> added on to them like
>>>
>>> + theme_minimal()
>>> + scale_colour_manual(values = c(“blue”, “red”, “gray50”)
>>> + etc.
>>>
>>> See the examples at
>>>
>>> http://docs.ggplot2.org/dev/vignettes/themes.html
>>>
>>> Especially for teaching, I find this to be a real problem. I don’t want to
>>> clutter up early examples with all this sort of formatting, but I do want to
>>> control the choice of colors used so that they work well for printing or
>>> projecting in my local environment. When I’m producing a journal or some
>>> other sort of report, this is less important because I generally don’t show
>>> the code in the final document — only the plot. But it still violates the
>>> DRY principle (Don’t Repeat Yourself).
>>>
>>> —rjp
>>>
>>> PS. I should have mentioned one commonly held, but poorly justified, reason
>>> to prefer ggplot2 over lattice: the default colors. lattice makes it easy
>>> to adjust the defaults. latticeExtra even provides a ggplot2 look-alike
>>> theme that will make your lattice plots look a lot like ggplot2 plots. (The
>>> won’t match exactly because they don’t include all of the same elements.
>>> Liking elements available in one system but not in the other is a possible
>>> reason to prefer one over the other.) On the other hand, if you don’t like
>>> the defaults in ggplot2 — you’re out of luck, you don’t get the change the
>>> defaults.
>
>
[[alternative HTML version deleted]]
More information about the R-sig-teaching
mailing list