ggplot2热图中相同数值显示不同颜色的问题排查

问题描述 #

使用ggplot2绘制混淆矩阵热图时，geom_text标注的数值显示正确，但相同数值对应的色块颜色却不一致（比如Urban行的0.04）。数据集及绘图代码如下：

数据集前8行 #

> head(zoo_cm.df, 8)
           Prediction      Reference Freq
1      Castle/Castrum Castle/Castrum 0.75
2           Religious Castle/Castrum 0.00
3               Rural Castle/Castrum 0.05
4   Rural site, villa Castle/Castrum 0.00
5               Urban Castle/Castrum 0.04
6 Urban, amphitheater Castle/Castrum 0.00
7      Castle/Castrum      Religious 0.00
8           Religious      Religious 1.00

数据集结构 #

> dput(zoo_cm.df)
structure(list(Prediction = structure(c(1L, 2L, 3L, 4L, 5L, 6L, 
1L, 2L, 3L, 4L, 5L, 6L, 1L, 2L, 3L, 4L, 5L, 6L, 1L, 2L, 3L, 4L, 
5L, 6L, 1L, 2L, 3L, 4L, 5L, 6L, 1L, 2L, 3L, 4L, 5L, 6L), levels = c("Castle/Castrum", 
"Religious", "Rural", "Rural site, villa", "Urban", "Urban, amphitheater"
), class = "factor"), Reference = structure(c(1L, 1L, 1L, 1L, 
1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 3L, 3L, 3L, 3L, 3L, 3L, 4L, 4L, 
4L, 4L, 4L, 4L, 5L, 5L, 5L, 5L, 5L, 5L, 6L, 6L, 6L, 6L, 6L, 6L
), levels = c("Castle/Castrum", "Religious", "Rural", "Rural site, villa", 
"Urban", "Urban, amphitheater"), class = "factor"), Freq = c(0.75, 
0, 0.05, 0, 0.04, 0, 0, 1, 0.05, 0, 0.03, 0, 0.17, 0, 0.74, 0, 
0.15, 0, 0, 0, 0.05, 0.33, 0.04, 0, 0.08, 0, 0.11, 0.67, 0.72, 
0, 0, 0, 0, 0, 0.01, 1)), row.names = c(NA, -36L), class = "data.frame")

原绘图代码 #

zoo_test_cm.plot <- ggplot(zoo_cm.df, aes(x=Prediction, y=Reference)) +
 geom_tile(aes(fill=Freq)) +
 geom_text(aes(Reference, Prediction, label = Freq), color = "black", size = 4) +
 scale_fill_gradient(low = "#FFFFFF", high = "#A6D1E6") +
  theme_minimal()+
  theme(axis.text.x = element_text(angle = 45, vjust = 1, 
    size = 10, hjust = 1), legend.position = "right")+
 coord_fixed()

问题根源 #

geom_text的映射参数错误：主图的aes定义了x=Prediction, y=Reference，但geom_text里却把x设为Reference、y设为Prediction，导致文本的位置与对应的色块完全错位。你看到的“相同数值颜色不同”，其实是文本标注到了错误的色块上，并非颜色刻度本身的问题。

比如Urban行的0.04，实际对应的色块是Prediction=Urban, Reference=Castle/Castrum，但文本却被放到了Prediction=Castle/Castrum, Reference=Urban的位置，而那个位置的Freq是0.08，所以颜色自然不一样。

修正后的代码 #

将geom_text的aes参数与主图保持一致，不需要重复指定x和y（因为会继承主图的映射），或者显式指定正确的x和y：

zoo_test_cm.plot <- ggplot(zoo_cm.df, aes(x=Prediction, y=Reference)) +
 geom_tile(aes(fill=Freq)) +
 # 方法1：继承主图的x/y映射，只指定label
 geom_text(aes(label = Freq), color = "black", size = 4) +
 # 方法2：显式指定正确的x/y（和主图一致）
 # geom_text(aes(x=Prediction, y=Reference, label = Freq), color = "black", size = 4) +
 scale_fill_gradient(low = "#FFFFFF", high = "#A6D1E6") +
 theme_minimal()+
 theme(axis.text.x = element_text(angle = 45, vjust = 1, 
    size = 10, hjust = 1), legend.position = "right")+
 coord_fixed()

验证效果 #

修正后，每个文本标注都会精准对应到所属的色块，数值和颜色刻度将完全匹配，相同数值的色块颜色也会一致。

内容的提问来源于stack exchange，提问作者Roberto