r - ggplot2中的中心对齐图例标题和图例键以获取长图例标题

Question

当图例标题很长时，我很难使图例的标题相对于图例键居中对齐。一年前有一个问题适用于短标题，但似乎不适用于长标题。

示例，首先使用简短的图例标题：

library(ggplot2)
ggplot(iris, aes(x=Sepal.Length, y=Sepal.Width, color=Petal.Width)) + geom_point(size = 3) +
  scale_color_distiller(palette = "YlGn", type = "seq", direction = -1,
                        name = "A") +
  theme(legend.title.align = 0.5)

一切如预期，图例标题在图例键上方居中。

现在与长图例标题相同：

ggplot(iris, aes(x=Sepal.Length, y=Sepal.Width, color=Petal.Width)) + geom_point(size = 3) +
  scale_color_distiller(palette = "YlGn", type = "seq", direction = -1,
                        name = "Long legend heading\nShould be centered") +
  theme(legend.title.align = 0.5)

我们可以看到文本与自身居中对齐，但与图例键无关。我曾尝试修改其他主题选项，例如legend.justification = "center"，但似乎没有一个将键从图例框中的最左侧位置移动。

一些评论：

• 几天前我正在运行ggplot2的开发版本，v2.2.1.9000。

• 我特别需要一个连续色阶调色板的解决方案。

Answer 1

2019 年 10 月 4 日更新：

不久前，我根据大约两年前在这里发布的原始想法编写了一个相当通用的函数。该函数在 github 上，但它不是任何官方发布的包的一部分。定义如下：

align_legend <- function(p, hjust = 0.5)
{
  # extract legend
  g <- cowplot::plot_to_gtable(p)
  grobs <- g$grobs
  legend_index <- which(sapply(grobs, function(x) x$name) == "guide-box")
  legend <- grobs[[legend_index]]

  # extract guides table
  guides_index <- which(sapply(legend$grobs, function(x) x$name) == "layout")

  # there can be multiple guides within one legend box  
  for (gi in guides_index) {
    guides <- legend$grobs[[gi]]

    # add extra column for spacing
    # guides$width[5] is the extra spacing from the end of the legend text
    # to the end of the legend title. If we instead distribute it by `hjust:(1-hjust)` on
    # both sides, we get an aligned legend
    spacing <- guides$width[5]
    guides <- gtable::gtable_add_cols(guides, hjust*spacing, 1)
    guides$widths[6] <- (1-hjust)*spacing
    title_index <- guides$layout$name == "title"
    guides$layout$l[title_index] <- 2

    # reconstruct guides and write back
    legend$grobs[[gi]] <- guides
  }

  # reconstruct legend and write back
  g$grobs[[legend_index]] <- legend
  g
}

该功能非常灵活和通用。以下是一些如何使用它的示例：

library(ggplot2)
library(cowplot)
#> 
#> ********************************************************
#> Note: As of version 1.0.0, cowplot does not change the
#>   default ggplot2 theme anymore. To recover the previous
#>   behavior, execute:
#>   theme_set(theme_cowplot())
#> ********************************************************
library(colorspace)

# single legend
p <- ggplot(iris, aes(Sepal.Width, Sepal.Length, color = Petal.Width)) + geom_point()
ggdraw(align_legend(p)) # centered

ggdraw(align_legend(p, hjust = 1)) # right aligned

# multiple legends
p2 <- ggplot(mtcars, aes(disp, mpg, fill = hp, shape = factor(cyl), size = wt)) + 
   geom_point(color = "white") +
   scale_shape_manual(values = c(23, 24, 21), name = "cylinders") +
   scale_fill_continuous_sequential(palette = "Emrld", name = "power (hp)", breaks = c(100, 200, 300)) +
   xlab("displacement (cu. in.)") +
   ylab("fuel efficiency (mpg)") +
   guides(
     shape = guide_legend(override.aes = list(size = 4, fill = "#329D84")),
     size = guide_legend(
       override.aes = list(shape = 21, fill = "#329D84"),
       title = "weight (1000 lbs)")
     ) +
   theme_half_open() + background_grid()

# works but maybe not the expected result
ggdraw(align_legend(p2))

# more sensible layout
ggdraw(align_legend(p2 + theme(legend.position = "top", legend.direction = "vertical")))

^{由reprex 包（v0.3.0）于 2019 年 10 月 4 日创建}

原答案：

我找到了解决方案。它需要对 grob 树进行一些挖掘，如果有多个图例，它可能不起作用，但否则这似乎是一个合理的解决方案，直到出现更好的解决方案。

library(ggplot2)
library(gtable)
library(grid)

p <- ggplot(iris, aes(x=Sepal.Length, y=Sepal.Width, color=Petal.Width)) + 
  geom_point(size = 3) +
  scale_color_distiller(palette = "YlGn", type = "seq", direction = -1,
                        name = "Long legend heading\nShould be centered") +
  theme(legend.title.align = 0.5)

# extract legend
g <- ggplotGrob(p)
grobs <- g$grobs
legend_index <- which(sapply(grobs, function(x) x$name) == "guide-box")
legend <- grobs[[legend_index]]

# extract guides table
guides_index <- which(sapply(legend$grobs, function(x) x$name) == "layout")
guides <- legend$grobs[[guides_index]]

# add extra column for spacing
# guides$width[5] is the extra spacing from the end of the legend text
# to the end of the legend title. If we instead distribute it 50:50 on
# both sides, we get a centered legend
guides <- gtable_add_cols(guides, 0.5*guides$width[5], 1)
guides$widths[6] <- guides$widths[2]
title_index <- guides$layout$name == "title"
guides$layout$l[title_index] <- 2

# reconstruct legend and write back
legend$grobs[[guides_index]] <- guides
g$grobs[[legend_index]] <- legend

grid.newpage()
grid.draw(g)

Answer 2

我破解了类似于 baptiste 在上述评论之一中描述的方式的源代码：将颜色条/标签/刻度 grobs 放入子 gtable 中，并将其定位为具有相同的行跨度/列跨度（取决于图例的方向）作为标题。

它仍然是一种 hack，但我想将其视为“整个会话一次 hack”的方法，而不必为每个情节手动重复这些步骤。

不同标题宽度/标题位置/图例方向的演示：

plot.demo <- function(title.width = 20,
                      title.position = "top",
                      legend.direction = "vertical"){
  ggplot(iris, 
         aes(x=Sepal.Length, y=Sepal.Width, color=Petal.Width)) + 
    geom_point(size = 3) +
    scale_color_distiller(palette = "YlGn",
                          name = stringr::str_wrap("Long legend heading should be centered",
                                                   width = title.width), 
                          guide = guide_colourbar(title.position = title.position),
                          direction = -1) +
    theme(legend.title.align = 0.5,
          legend.direction = legend.direction)
}

cowplot::plot_grid(plot.demo(),
                   plot.demo(title.position = "left"),
                   plot.demo(title.position = "bottom"),
                   plot.demo(title.width = 10, title.position = "right"),
                   plot.demo(title.width = 50, legend.direction = "horizontal"),
                   plot.demo(title.width = 10, legend.direction = "horizontal"),
                   ncol = 2)

这也适用于多个颜色条图例：

ggplot(iris, 
       aes(x=Sepal.Length, y=Sepal.Width, 
           color=Petal.Width, fill = Petal.Width)) + 
  geom_point(size = 3, shape = 21) +
  scale_color_distiller(palette = "YlGn",
                        name = stringr::str_wrap("Long legend heading should be centered",
                                                 width = 20),
                        guide = guide_colourbar(title.position = "top"),
                        direction = -1) +
  scale_fill_distiller(palette = "RdYlBu",
                       name = stringr::str_wrap("A different heading of different length",
                                                width = 40),
                       direction = 1) +
  theme(legend.title.align = 0.5,
        legend.direction = "vertical",
        legend.box.just = "center")

（旁注：legend.box.just = "center"需要正确对齐两个图例。我担心了一段时间，因为目前只有“top”、“bottom”、“left”和“right”被列为可接受的参数值，但事实证明两者“中心”/“中心”也被底层接受grid::valid.just。我不确定为什么?theme帮助文件中没有明确提到这一点；尽管如此，它确实有效。）

要更改源代码，请运行：

trace(ggplot2:::guide_gengrob.colorbar, edit = TRUE)

并从此更改最后一段代码：

  gt <- gtable(widths = unit(widths, "cm"), heights = unit(heights, 
    "cm"))
  ... # omitted
  gt
}

对此：

  # create legend gtable & add background / legend title grobs as before (this part is unchanged)
  gt <- gtable(widths = unit(widths, "cm"), heights = unit(heights, "cm"))
  gt <- gtable_add_grob(gt, grob.background, name = "background", 
                        clip = "off", t = 1, r = -1, b = -1, l = 1)
  gt <- gtable_add_grob(gt, justify_grobs(grob.title, hjust = title.hjust, 
                                          vjust = title.vjust, int_angle = title.theme$angle, 
                                          debug = title.theme$debug), name = "title", clip = "off", 
                        t = 1 + min(vps$title.row), r = 1 + max(vps$title.col), 
                        b = 1 + max(vps$title.row), l = 1 + min(vps$title.col))

  # create child gtable, using the same widths / heights as the original legend gtable
  gt2 <- gtable(widths = unit(widths[1 + seq.int(min(range(vps$bar.col, vps$label.col)), 
                                                 max(range(vps$bar.col, vps$label.col)))], "cm"),
                heights = unit(heights[1 + seq.int(min(range(vps$bar.row, vps$label.row)), 
                                                   max(range(vps$bar.row, vps$label.row)))], "cm"))

  # shift cell positions to start from 1
  vps2 <- vps[c("bar.row", "bar.col", "label.row", "label.col")]
  vps2[c("bar.row", "label.row")] <- lapply(vps2[c("bar.row", "label.row")],
                                            function(x) x - min(unlist(vps2[c("bar.row", "label.row")])) + 1)
  vps2[c("bar.col", "label.col")] <- lapply(vps2[c("bar.col", "label.col")],
                                            function(x) x - min(unlist(vps2[c("bar.col", "label.col")])) + 1)

  # add bar / ticks / labels grobs to child gtable
  gt2 <- gtable_add_grob(gt2, grob.bar, name = "bar", clip = "off",
                         t = min(vps2$bar.row), r = max(vps2$bar.col),
                         b = max(vps2$bar.row), l = min(vps2$bar.col))
  gt2 <- gtable_add_grob(gt2, grob.ticks, name = "ticks", clip = "off",
                         t = min(vps2$bar.row), r = max(vps2$bar.col),
                         b = max(vps2$bar.row), l = min(vps2$bar.col))
  gt2 <- gtable_add_grob(gt2, grob.label, name = "label", clip = "off",
                         t = min(vps2$label.row), r = max(vps2$label.col),
                         b = max(vps2$label.row), l = min(vps2$label.col))

  # add child gtable back to original legend gtable, taking tlrb reference from the
  # rowspan / colspan of the title grob if title grob spans multiple rows / columns.
  gt <- gtable_add_grob(gt, justify_grobs(gt2, hjust = title.hjust, 
                                          vjust = title.vjust), 
                        name = "bar.ticks.label", clip = "off", 
                        t = 1 + ifelse(length(vps$title.row) == 1, 
                                       min(vps$bar.row, vps$label.row),
                                       min(vps$title.row)), 
                        b = 1 + ifelse(length(vps$title.row) == 1, 
                                       max(vps$bar.row, vps$label.row),
                                       max(vps$title.row)), 
                        r = 1 + ifelse(length(vps$title.col) == 1, 
                                       min(vps$bar.col, vps$label.col),
                                       max(vps$title.col)), 
                        l = 1 + ifelse(length(vps$title.col) == 1, 
                                       max(vps$bar.col, vps$label.col),
                                       min(vps$title.col)))
  gt
}

要撤消更改，请运行：

untrace(ggplot2:::guide_gengrob.colorbar)

使用的包版本：ggplot2 3.2.1。

Answer 3

您必须更改源代码。目前它计算标题 grob 和 bar+labels 的宽度，并在视口（gtable）中左对齐 bar+labels 。这是硬编码的。