Add sym_limits function

[dylanhmorris]

Closes #30

[damonbayer]

Thanks @dylanhmorris!

It would be great if this could be implemented to more directly work with ggplot2, something like scale_y_symmetric. The LLM results I got based on this idea were pretty ugly looking, though.

[damonbayer]

@dylanhmorris what do you think about using scales to support a wider range of transformations in this function?

[damonbayer]

@dylanhmorris

sym_lim <- function(values, center = 0, transform = "identity") {
  transform_fn <- scales::as.transform(transform)
  transformed_values <- transform_fn$transform(values)
  transformed_center <- transform_fn$transform(center)
  span <- max(abs(transformed_values - transformed_center))
  
  transform_fn$inverse(transformed_center + c(-span, span))
}

sym_lim(values = c(3, 5))
#> [1] -5  5

sym_lim(values = c(0.4, 10), center = 1, transform = "log10")
#> [1]  0.1 10.0

sym_lim(values = c(2, 9), center = 4, transform = "sqrt")
#> [1] 1 9

sym_lim(values = c(1, 100), center = 1, transform = "log1p")
#> [1]  -0.960396 100.000000

^{Created on 2025-05-15 with reprex v2.1.1}

[dylanhmorris]

@dylanhmorris what do you think about using scales to support a wider range of transformations in this function?

I love this! PR it into this branch @damonbayer?

[dylanhmorris]

@damonbayer What do you think about making the center value default to NULL and using "0 on the transformed scale" as the center value if the user doesn't set it, but having the user specify center values on the untransformed scale?

That way you'd get sensible defaults and avoid problems when 0-on-the-untransformed scale is outside (or at the edge of) the domain of the transform, e.g.

sym_limits(c(5, 2), transform = "log10")

[damonbayer]

@damonbayer What do you think about making the center value default to NULL and using "0 on the transformed scale" as the center value if the user doesn't set it, but having the user specify center values on the untransformed scale?

I dig it. Great heuristic for getting the "natural" center of a scale. Let's swap the order of the center and transform arguments, then.

[dylanhmorris]

LMK how this looks @damonbayer. Feel free to merge if it looks good.