带有小提琴图的 R scale_y_facet：负对数值

Question

假设一个数据框

n <- 100
d <- data.frame("name" = c(rep("A", n), rep("B", n)),
                "type" = rep(c("X", "Y"), 2*n), 
                "value" = c(rnorm(n, 0, 1), rbeta(n, 1, 3)))

以及以下小提琴情节：

ggplot(d, aes(name, value, fill = type)) + 
  geom_violin(trim = F, position = dodge, na.rm = T) + 
  geom_boxplot(width = 0.1, position = dodge, na.rm = T) +
  facet_wrap(facets = ~name, scales = 'free') +
  theme(legend.position = "bottom",
        axis.title.x = element_blank(), 
        legend.title = element_blank(),
        axis.text.x = element_blank(),
        axis.ticks.x = element_blank(),
        legend.key.width = unit(0.5, "cm"))   +
  ggh4x::scale_y_facet(
    name == "B",
    trans  = "log10",
    breaks = breaks_log(),
    labels = label_log()) 

不幸的是，当我收到消息时，这不起作用

Warning messages:
1: In transformation$transform(x) : NaNs produced
2: In scale_facet(expr = enquo(expr), "y", ..., type = type) :
  log-10 transformation introduced infinite values.
3: In transformation$transform(x) : NaNs produced
4: In scale_facet(expr = enquo(expr), "y", ..., type = type) :
  log-10 transformation introduced infinite values.
5: In transformation$transform(x) : NaNs produced
6: In scale_facet(expr = enquo(expr), "y", ..., type = type) :
  log-10 transformation introduced infinite values.
7: In transformation$transform(x) : NaNs produced
8: In scale_facet(expr = enquo(expr), "y", ..., type = type) :
  log-10 transformation introduced infinite values.

输入对数变换的值严格为正，因此这不是问题。我认为问题在于，仅在对数变换之后我才会获得负值（其指标是 y 轴的限制从 0 以上开始）。所以，我需要调整 y 轴，或者更准确地说，允许

breaks

取负值。如何做到这一点？

Answer 1

小提琴图在极值处扩展了数据，因此您可能会得到一些负值。您可以使用 pseudo_log 转换来防止警告：

ggplot(d, aes(name, value, fill = type)) + 
  geom_violin(trim = F, position = "dodge", na.rm = T) + 
  geom_boxplot(width = 0.1, position = "dodge", na.rm = T) +
  facet_wrap(facets = ~name, scales = 'free') +
  theme(legend.position = "bottom",
        axis.title.x = element_blank(), 
        legend.title = element_blank(),
        axis.text.x = element_blank(),
        axis.ticks.x = element_blank(),
        legend.key.width = unit(0.5, "cm"))   +
  ggh4x::scale_y_facet(
    name == "B",
    #trans  = "log10",
    #breaks = breaks_log(),
    trans  = pseudo_log_trans(base = 10),
    breaks = 10^(c(-5, -1, 0)),
    labels = label_log()) 

请注意，如果您省略对

violin_plot

的调用，您将不会收到警告。