incremental adjustment demonstrations

[EvaMaeRey]

I wonder what people's favorite ways of showing adjustment consequences are. For me, last_plot() + new_code() was key to really 'getting' ggplot2. another strategy is p1 + new_code(). And there's also, quote complete code and highlight what's new e.g. codehover/flipbookr, others.

I put together ggplotXclosereads https://evamaerey.github.io/crispy-octo-train/closereads-chunk-reveal.html that uses last_plot() + new_code().

[teunbrand]

I can see last_plot() working well for slides where you incrementally build things. However, in real life (or, mine at least) I never get a component down perfectly the first time. So my process looks usually something like the following:

library(ggplot2)

ggplot(mpg, aes(displ, hwy)) +
  geom_point()

# Edit global aes
ggplot(mpg, aes(displ, hwy, colour = drv)) +
  geom_point()

# Display as filled points
ggplot(mpg, aes(displ, hwy, colour = drv)) +
  geom_point(shape = 21)

# Oops, now colour must be fill
ggplot(mpg, aes(displ, hwy, fill = drv)) +
  geom_point(shape = 21)

# Continue on
ggplot(mpg, aes(displ, hwy, fill = drv)) +
  geom_point(shape = 21) +
  facet_wrap(~ year)

I can't really last_plot() my way through this, especially if I'm making mistakes along the way. The issue I'd have if I were to teach somebody to use ggplot2, is that last_plot() gets conveyed as the preferred way to build a plot, whereas it really is more of a convenience function. In addition, in contrast to assigning a plot to some temporary variable (say p), you cannot really side-track in an interactive session without affecting the plot that is stored in the emphemoral last_plot().

[teunbrand]

Yeah I think what makes it into the theme are non-data decisions relating to graphical properties that one might want to tweak in sync with other parts of the theme. If you just need a one off change of point colour you can still just do geom_point(colour = "red") or whatever.

[EvaMaeRey]

I do kind of like the idea of extenders utilities repo in principle @mjskay, @yjunechoe. like a 'ggplot2 tricks' but for extenders, and it could maybe become a package or maybe it stays like @teunbrand's tricks and remains principally copy and paste resource.

modifyList for updating aes and params, but I encountered some corner case because modifyList modifies elements recursively...

@mjskay do you think ggplot2:::modify_list (from @yjunechoe) would correctly handle the principle and corner cases, by the way?

ggplot2:::modify_list
#> function (old, new) 
#> {
#>     for (i in names(new)) old[[i]] <- new[[i]]
#>     old
#> }

modifyList definitely looks it's trying to do a lot more (but maybe more than is needed)...

function (x, val, keep.null = FALSE) 
{
    stopifnot(is.list(x), is.list(val))
    xnames <- names(x)
    vnames <- names(val)
    vnames <- vnames[nzchar(vnames)]
    if (keep.null) {
        for (v in vnames) {
            x[v] <- if (v %in% xnames && is.list(x[[v]]) && is.list(val[[v]])) 
                list(modifyList(x[[v]], val[[v]], keep.null = keep.null))
            else val[v]
        }
    }
    else {
        for (v in vnames) {
            x[[v]] <- if (v %in% xnames && is.list(x[[v]]) && 
                is.list(val[[v]])) 
                modifyList(x[[v]], val[[v]], keep.null = keep.null)
            else val[[v]]
        }
    }
    x
}

A modified version of x, with the modifications determined as follows (here, list elements are identified by their names). Elements in val which are missing from x are added to x. For elements that are common to both but are not both lists themselves, the component in x is replaced (or possibly deleted, depending on the value of keep.null) by the one in val. For common elements that are in both lists, x[[name]] is replaced by modifyList(x[[name]], val[[name]]).

I initially tried modifyList() here but switched to a c.uneval() method since this use case should not involve overwriting...but i'm waiting for a big unanticipated problem with this trick to arise.

@corybrunson sounds interesting. Don't have my mind wrapped around this. 🤷‍♀️

[corybrunson]

sounds interesting. Don't have my mind wrapped around this. 🤷‍♀️

Basically, the uneval class does not forbid the duplication of names, but one can manually assign the class to a list with duplicated names, and c.uneval() as written does not forbid this. This is just because it doesn't expect it; maybe it should be forbidden, but in that case c() is the wrong generic for the job. (I'm convincing myself that i should replace it with the modify_list() above.)

[corybrunson]

The (first) unanticipated problem turned out to be that {ggplot2} itself concatenates aesthetics internally and relies on c() dropping their class and returning a list with possible duplicate names, e.g. here. I think this renders c.uneval() wholly inappropriate for extensions at this point. (I'm trying the same function with the new name aes_c() instead, which is still problematic, as it throws errors when names are duplicated. But i don't want users to have to type modify_aes() in a ggplot() call, for example, since it suggests something other than the normal use case.)

[EvaMaeRey]

keep.null is nice from modifyList in the following context (geom_textpath inspired), else NULL elements are dropped.

library(ggplot2); 

GeomText$default_aes |> 
   modifyList(GeomLine$default_aes) |> 
   modifyList(aes(spacing = 0, linecolor = NULL, textcolor = NULL), 
                       keep.null = T)

#> Aesthetic mapping: 
#> * `colour`     -> `from_theme(ink)`
#> * `family`     -> `from_theme(family)`
#> * `size`       -> `from_theme(fontsize)`
#> * `angle`      -> 0
#> * `hjust`      -> 0.5
#> * `vjust`      -> 0.5
#> * `alpha`      -> NA
#> * `fontface`   -> 1
#> * `lineheight` -> 1.2
#> * `linewidth`  -> `from_theme(linewidth)`
#> * `linetype`   -> `from_theme(linetype)`
#> * `spacing`    -> 0
#> * `linecolour` -> NULL
#> * `textcolour` -> NULL

^{Created on 2025-06-05 with reprex v2.1.1}

https://github.com/AllanCameron/geomtextpath/blob/cb673153b6349a791bf8434e8fe954b332b62ee0/R/geom_textpath.R#L184C17-L199C4

Current specification:

  default_aes = aes(
    colour     = "black",
    size       = 3.88,
    hjust      = 0.5,
    vjust      = 0.5,
    family     = "",
    fontface   = 1,
    lineheight = 1.2,
    alpha      = NA,
    linewidth  = 0.5,
    linetype   = 1,
    spacing    = 0,
    linecolour = NULL,
    textcolour = NULL,
    angle      = 0
  ),

[EvaMaeRey]

@corybrunson, was looking at your ggalluvial with this closereads and decomposed code approach to try to capture a data exploration. Playing around with a in-script geom_stratum_text() definition - I still think that is the way to go (and definitely adding the default aes makes sense to me)! 🙃

Anyway, package is really fun to use. https://evamaerey.github.io/crispy-octo-train/ggalluvial.html you have to scroll a bit to get to the 'scrollytelling' part...

[corybrunson]

What's the "eyes" reaction meant to convey? I would use it here to indicate that i'm taking a closer look at this than allows for a settled response.

[EvaMaeRey]

My go-to reaction is the eyes. I think it means 'interesting!'

[EvaMaeRey]

Just seeing this issue of @willgearty's related to geom defaults.... willgearty/deeptime#77

[EvaMaeRey]

And ...

GeomPointFit <- ggproto("GeomPointFit", GeomPoint, 
              default_aes = modifyList(GeomPoint$default_aes, 
                                        GeomSmooth$default_aes[c("colour")]))

To combine with StatSmoothFit and StatSmoothResiduals (or similar, e.g. ggpmisc::StatFitDeviations) where xseq = data$x. Using, GeomSmooth$default_aes[c("colour")]) since it might be good to have the model results w same color scheme as GeomSmooth

[EvaMaeRey]

kind of notes to self here, but last thing, I keep thinking about letting the accent be accessible to the user in theme_grey() and the like. @teunbrand 🎈🙏 I think geom_smooth plays such a large role in ggplot2 education, that this merit's thinking about this one more time. Like the go-to for showing how you can layer up geom_* or stat_* is geom_point() + geom_smooth(). So accent is seen frequently even though it's only shows up in a handful of functions.

[teunbrand]

You can change the accent colour using theme(geom = element_geom(accent = ...)). I don't see an added benefit by allowing e.g. theme_gray(accent = ...): it is overly specific and uncoordinated with any other settings. If you want to attempt to convince other ggplot2 maintainers of the merit of accent, or gauge interest in that setting in a slightly wider audience, you're welcome to file an issue.

[EvaMaeRey]

Okay. I think I will do that. Thanks!

[EvaMaeRey]

tidyverse/ggplot2#6438 🎉