使用预先创建的密度值创建 geom_violin 图

Question

我已经在不同的因子水平上创建了一些加权核密度估计，我认为这些估计不能合并到 geom_violin 图估计中。我想知道是否有一种方法 geom_violin 或其他 ggplot2 函数可以使用原始数据来创建小提琴图而不是内置密度计算？任何帮助将非常感激。一些示例代码，我想根据 y 值在 x 值分布上的变化创建小提琴图...

###Create data
Surge_vs_Plummet_Stats_df <- data.frame(
Type = rep(c("Surge", "Plummet"), each = 50),
site_no = sample(1:10, 100, replace = TRUE),
Mean = c(rnorm(50, mean = 5, sd = 2), rnorm(50, mean = 3, sd = 1)))

###Calculate weights
station_counts <- table(Surge_vs_Plummet_Stats_df$site_no)
Surge_vs_Plummet_Stats_df$Weights <- 1 / station_counts[Surge_vs_Plummet_Stats_df$site_no]
Surge_vs_Plummet_Stats_df$Weights <- Surge_vs_Plummet_Stats_df$Weights / 
sum(Surge_vs_Plummet_Stats_df$Weights)##Normalize (sum to 1)

###Identify bandwidth
bw <- bw.nrd(Surge_vs_Plummet_Stats_df$Mean)##Not 100% sure its doing much

###Now separate the dfs run KDEs
Surge_vs_Plummet_Stats_Surge <- Surge_vs_Plummet_Stats_df%>%filter(Type == "Surge")%>%mutate(Weights = Weights / sum(Weights))
Surge_kde <- density(Surge_vs_Plummet_Stats_Surge$Mean, weights = Surge_vs_Plummet_Stats_Surge$Weights,bw = bw,
                 from=min(Surge_vs_Plummet_Stats_df$Mean), to=max(Surge_vs_Plummet_Stats_df$Mean))##Delib the full df and not just surges
#
Surge_vs_Plummet_Stats_Plummet <- Surge_vs_Plummet_Stats_df%>%filter(Type == "Plummet")%>%mutate(Weights = Weights / sum(Weights))
Plummet_kde <- density(Surge_vs_Plummet_Stats_Plummet$Mean, weights = Surge_vs_Plummet_Stats_Plummet$Weights,bw = bw,
                   from=min(Surge_vs_Plummet_Stats_df$Mean), to=max(Surge_vs_Plummet_Stats_df$Mean))##Delib the full df and not just surges
##
Mean_Kernel_df <- data.frame(x = c(Surge_kde$x,Plummet_kde$x),
                         y = c(Surge_kde$y,Plummet_kde$y),
                         Type = c(rep("Surge",times=length(Surge_kde$x)),
                                  rep("Plummet",times=length(Surge_kde$y))))

Answer 1

您通常可以通过使用

stat_

来使用没有相关统计变换（即它们的

stat = "identity"

）的几何图形。我们也可以对

geom_violin

执行此操作，如下所示：

ggplot(Mean_Kernel_df, aes(Type, x, violinwidth = y)) + 
  geom_violin(stat = 'identity')

但是，这会产生一些警告，并且使用

violinwidth

作为美学并没有记录在案，因此这在将来可能会被打破。