`tinyplot` method(s)?

[grantmcdermott]

This might be encroaching on see's toes, but is there any possibility of supporting tinyplot methods for parameters? It would only require a lightweight Suggests addition and I think we could keep the implementation pretty simple. Something along the lines of:

tinyplot.parameters_model = function(
  x,
  flip = TRUE,
  zero = TRUE,
  ...
) {
  xlabs = attr(x, "pretty_labels")
  x$Parameter = xlabs[x$Parameter]
  tinyplot(
    Coefficient ~ Parameter, ymin = CI_low, ymax = CI_high,
    data = x,
    type = "pointrange",
    flip = flip,
    ...
  )
  if (zero) if (flip) abline(v = 0, lty = 2) else abline(h = 0, lty = 2)
}

MWE:

library(tinyplot)
library(parameters)

res = parameters(lm(wt ~ am * cyl, data = mtcars))
plt(res)

Slightly fancier version with a theme:

tinytheme("classic")
plt(res)

^{Created on 2025-06-04 with reprex v2.1.1}

Happy to iterate on a potential PR if there's interest on your side, but no offence if it's out of scope.

[strengejacke]

Sure! I was indeed thinking already about this, also regarding a broader support.

Maybe even support for visualization recipes. Basically, this sets up a layer with ggplot geoms/aes, which is more or less simply plotted using geoms_from_list(). This approach has the advantage that we can have the plot preparation inside the related packages, without adding dependencies, and the see package is only needed for a simple plot method. But that's probably an overkill for now.

I'm happy to have tinyplot methods for the easystats packages, especially modelbased. Maybe we can even use it as alternative backend for some performance plots.

[strengejacke]

For model_parameters(), also consider models with multiple components (zero-inflation) or group levels of random effects. The full overview of possible plot-support is in the related vignette from the see package: https://easystats.github.io/see/articles/parameters.html

[grantmcdermott]

Great. I'm a little pressed for time over the next two weeks, but wanted to sound out your interest so we could brainstorm a bit. Looking over those examples quickly, I think we could support most of the plot types out of the box. Getting the performance grid plot might be a little trickier because of legend restrictions (https://github.com/grantmcdermott/tinyplot/issues/329), but we could think about potential workarounds.

[strengejacke]

Sounds good! I don't think we have to replicate plots 1:1. If there are alternative layout or styles, e.g. for check_model(), that's ok, too. But the performance package is maybe something that could be addressed later, modelbased is probably easier and maybe "more important", as those are the plots you probably put into presentations/publications, similar to parameters. Or those posterior distribution plots we have in bayestestR would be nice to have as methods for tinyplot.