override.aes与list参数的命名向量

Question

As far as I can see, when using override.aes , one needs to "hard code" values in the list argument depending on the order of the legend... Is it possible to used named vectors instead?

Example:

library(ggplot2)
set.seed(1)
df1 <- data.frame(x = 1:25, y_line = rnorm(25, 1, 0.01))

p <- ggplot(df1, aes(x, y_line)) +
      geom_line(aes(colour = 'line')) +
      geom_point(aes(y = 1, colour = 'point')) + 
      geom_line(aes(y = 1, color = 'point'), linetype = 'dotted') 

# This specifies values by order:
p + guides(colour = guide_legend(override.aes = list(linetype = c('dotted', 'solid'), shape = c(NA, 16))))

# Attempt to use named vectors does not change the output 
p + guides(colour = guide_legend(override.aes = list(linetype = c(point = 'dotted', line = 'solid'), shape = c(NA, 16))))

Both result in the same legend:

This is the same when using quoted names.

^{Created on 2019-05-07 by the reprex package (v0.2.1)}

Answer 1

I'm sorry but I think I may have missed exactly for what purpose you would want to use this, so I apologize if my workaround doesn't adress your need directly.

I'm unsure wether it can be done within the guides function, but what you can do, is to use guide_legend() in conjunction with scale_colour_* to set the order in advance with breaks . I think this mimicks your intended behaviour:

p <- ggplot(df1, aes(x, y_line)) +
  geom_line(aes(colour = 'line')) +
  geom_point(aes(y = 1, colour = 'point')) +
  geom_line(aes(y = 1, colour = 'point'), linetype = 'dotted') +
  scale_colour_discrete(
    breaks = c("point", "line"),
    guide = guide_legend(override.aes = list(linetype = c("dotted", "solid"),
                                             shape = c(16, NA)))
    )

Answer 2

Note: I haven't tested this out thoroughly, & there will probably be unforeseen hiccups down the line when the hack interacts with other unobserved functions within ggplot2. The package's inner workings can be pretty hard to divine, but hopefully this provides a starting point...

The legend building part occurs within ggplot2:::build_guides (un-exported function). As you've observed, the names of a named vector in override.aes are ignored during the process. One possible workaround is to insert some code into the function to get the named vector in the correct (based on legend labels) order. I've also added a check for the default aesthetic parameters, for cases where we may want to override the aesthetics for only one or two labels, & leave the rest as default.

Here's the code to be inserted. I've only tried it on linetype, shape, & size. Offhand, linetype is the only case that comes to my mind with both numerical & categorical values, so that's the specific scenario covered below for default.aes .

# define a function that completes each element in the override.aes list if
# it's a named vector, by arranging it in the order used by the legend labels,
# & replacing any unsupplied value with the latest (based on most recent layer) 
# default aesthetic value for that specific element
complete.override.aes <- function(gdef, default.aes){
  override.aes <- gdef$override.aes
  if(!any(sapply(override.aes, function(x) !is.null(names(x))))){
    return(gdef)
  }
  key.label <- gdef$key$.label
  for(i in seq_along(override.aes)){
    if(!is.null(names(override.aes[[i]]))){
      x <- override.aes[[i]][key.label]
      default.x <- default.aes[[names(override.aes)[[i]]]]
      if(!is.na(default.x)){
        x <- dplyr::coalesce(x,
                             rep(default.x,
                                 times = length(key.label)))
      }
      names(x) <- NULL
      override.aes[[i]] <- x
    }
  }
  gdef$override.aes <- override.aes
  gdef
}

# extract default aes associated with each layer in ggplot object,
# combine, & remove duplicates (keep latest where applicable)
default.aes <- sapply(layers, function(x) x$geom$default_aes)
default.aes <- purrr::flatten(default.aes)
default.aes <- default.aes[!duplicated(default.aes, fromLast = TRUE)]
# for linetype (if applicable), map from numeric to string
if(!is.null(default.aes[["linetype"]]) &
   is.numeric(default.aes[["linetype"]])){
  if(default.aes[["linetype"]] == 0) default.aes[["linetype"]] <- 7
  default.aes[["linetype"]] <- c("solid", "dashed", "dotted",
                                 "dotdash", "longdash", "twodash",
                                 "blank")[default.aes[["linetype"]]]
}

gdefs <- lapply(gdefs, complete.override.aes, default.aes)

To use this, run trace(ggplot2:::build_guides, edit = TRUE) and insert the above code after line 32 (ie after return(zeroGrob()) & before gdefs <- guides_merge(gdefs) ).

(Alternatively, we can insert the code into our own version of the above function & name it build_guides2 , define a modified version of ggplot2:::ggplot_gtable.ggplot_built which calls on that instead of ggplot2:::build_guides , then a modified version of ggplot2:::print.ggplot which calls on that instead of ggplot_gtable . However, this gets unwieldy quickly, takes up a considerable amount of space, & is tangential to the topic at hand, so I'm not going into details for that here.)

Result:

# correct mapping for linetype
p + guides(colour = guide_legend(
             override.aes = list(linetype = c(point = 'dotted', line = 'solid'), 
                                 shape = c(NA, 16))))

# both linetype & shape use named vectors, & specify one value each
# (otherwise linetype defaults to "solid" & shape to 19)
p + guides(colour = guide_legend(
             override.aes = list(linetype = c(point = 'dotted'), 
                                 shape = c(line = 8))))