Interaction between breaks and scales is confusing

[wch]

This is a followup to #264.

p <- qplot(mpg, wt, data = mtcars, colour = factor(cyl))

# Works (fixed in 264)
p + scale_colour_manual(values = c("8" = "red","4" = "blue","6" = "green"))

# Doesn't work
p + scale_colour_manual(breaks=c(8, 4, 6), values=c("red", "green", "blue"))

This should be added to the existing test:

test_that("names of values used in manual scales", {
  # ... other test already here...

   s <- scale_colour_manual(breaks = c(8, 4, 6), values = c("c", "a", "b"))
   scale_train(s, c("4", "6", "8"))
   expect_equal(scale_map(s, c("4", "6", "8")), c("a", "b", "c"))  
})

[kohske]

p + scale_colour_manual(limits=c(8, 4, 6), values=c("red", "green", "blue"))

works.

p + scale_colour_manual(limits=c(8, 4, 6), values=c("red", "green", "blue"), breaks = c(4,8))

also works.

p + scale_colour_manual(limits=c(8, 4), values=c("red", "green"), breaks = c(4,8))

in this case cyl=6 does not appear as expected.

So the general rule is:

• limits determines if the corresponding data appears in panel. With discrete scale it also determines the (internal) order of the variable.
• breaks control guides.

I agree that this may be confusing though.

[wch]

Do you mean that it works in your branch? It doesn't work on master for me.

Update: oh, nevermind, it does work. I was confused. The limits/breaks thing trips me up sometimes.

[kohske]

It works on master branch 5e89ea1

[kohske]

And now I'd think that the best way to control the order of discrete scale is to specify limits explicitly:

p <- qplot(factor(1:5, labels = letters[5:1]), 1:5) + geom_point()
p
p + scale_x_discrete(limits = letters[1:5])
p + scale_x_discrete(limits = letters[c(1, 3, 2, 4, 5)])

This is easier than reordering the factor of data.

This may be documented.

limits: for continuous scale, 2 length vector of upper and lower bound.
for discrete scale, the levels that appear in the plot (I think this is same as continuous). Furthermore, the order of limits affects the order of the variable in the panel and legend.

[kohske]

@wch could you please document this?

[wch]

Sure... do you think it should go in scale_x_discrete?

[kohske]

Thanks!

I'd think we may have a group document for scale like ?Arithmetic or ?Comparison, and each help of scale_X_Y may have see also scale.
I'm not sure how we can do this by roxygen though...

[kohske]

And also it would be better if we have a grouped help page such as ?coord, ?stat, ?geom etc that illustrate how and what each component do in general with the list of available children (ie, coord_cartesian, coord_polar, etc).

I'm not good at writing document, and I imagine this will include a large work so milestone 0.9.2 or 1.0 is appropriate?

[kohske]

there is already ggscale for help document, see scale-.r

probably it can be improved:

#' scales in ggplot2
#'
#' Scales are an S3 class with a single mutable component implemented with
#' a reference class - the range of the data.  This mutability makes working
#' with scales much easier, because it makes it possible to distribute the
#' training, without having to worry about collecting all the pieces back 
#' together again.
#'
#' \section{Components of scale}
#'
#' \enumerate{
#'  \item limits
#'  \item breaks
#'  \item labels
#'  \item guide
#' }
#'
#' @name ggscale
#' @S3method print scale
#' @family scales
NULL

all scale_x_y will refer this topic.

[hadley]

Moving to 0.9.2 since this is a documentation issue

[wch]

I pushed a change that fixes the original issue. The other issues that we discussed - maybe they should be filed as a new issue.