Feature request: rescale transformation

[teunbrand]

I'd like to propose a new transformation function that essentially wraps the rescale() function. It's definition can be something like the following:

library(scales)

transform_rescale <- function(to = c(0, 1), from = c(0, 1)) {
  force(to)
  force(from)
  new_transform(
    "rescaling",
    transform = function(x) rescale(x, to = to, from = from),
    inverse   = function(x) rescale(x, to = from, from = to) 
  )
}

It should be able to handle a bunch of 'scale and translate' transformations, like unit conversions. Going from metres to inches is simply a scale transformation.

metres_to_inches <- transform_rescale(to = c(0, 100/2.54))
metres_to_inches$transform(1)
#> [1] 39.37008

Whereas going from degrees celcius to degrees fahrenheit is a 'scale and translate' transformation.

celsius_to_fahrenheit <- transform_rescale(to = c(32, 212), from = c(0, 100))
celsius_to_fahrenheit$transform(-5)
#> [1] 23
celsius_to_fahrenheit$transform(32)
#> [1] 89.6

^{Created on 2023-12-15 with reprex v2.0.2}

Given that it is suitable for plenty of unit conversions, I think this transformation may have a bunch of applications.

[mjskay]

In case it is desired, adding derivative support for this should entail something like:

transform_rescale <- function(to = c(0, 1), from = c(0, 1)) {
  force(to)
  force(from)
  new_transform(
    "rescaling",
    transform = function(x) rescale(x, to = to, from = from),
    inverse   = function(x) rescale(x, to = from, from = to),
    d_transform = function(x) rep(diff(to) / diff(from), length(x)),
    d_inverse = function(x) rep(diff(from) / diff(to), length(x))
  )
}

[teunbrand]

Ah yes, of course! Thanks for weighing in, it was a little oversight on my part

[thomasp85]

Happy to accept a PR for this :-)

[teunbrand]

There was something fundamentally flawed with this approach when I last tried it. Unfortunately, I have forgotten what it was, so let me try again :)

[teunbrand]

I've rediscovered why this approach is flawed. Essentially, you calculate breaks and labels in 'original' space, so the transformation is essentially a fancyful transform_identity() in disguise.

library(scales)

transform_rescale <- function(to = c(0, 1), from = c(0, 1)) {
  scales:::force_all(to, from)
  new_transform(
    "rescaling",
    transform   = function(x) rescale(x, to = to, from = from),
    inverse     = function(x) rescale(x, to = from, from = to),
    d_transform = function(x) rep(diff(to) / diff(from), length(x)),
    d_inverse   = function(x) rep(diff(from) / diff(to), length(x))
  )
}

celcius_to_fahrenheit <- transform_rescale(to = c(32, 212), from = c(0, 100))

library(ggplot2)

# Note the original range in Celcius:
range(pressure$temperature)
#> [1]   0 360

ggplot(pressure, aes(temperature, pressure)) +
  geom_line() +
  scale_x_continuous(
    transform = celcius_to_fahrenheit
  )

^{Created on 2024-10-22 with reprex v2.1.1}

It might be easy enough to calculate the default breaks in the transformed space, but that won't mesh well if people give custom breaks in ggplot2.

I'll just go ahead and close this.