ggplot2中geom_hexbin分箱计数与渐变填充颜色映射不准确问题

我尝试使用ggplot2绘制分箱散点图，基础代码如下：

library(ggplot2)
bks = seq(from = 0, to = 10000, by = 1000)
d <- ggplot(diamonds, aes(carat, price)) + theme_bw()
d + geom_point(alpha = 0.01)

使用geom_hex时，分箱计数与渐变刻度的映射不准确：

d + geom_hex(aes(fill = after_stat(count)), bins = 30, colour = "white") + 
  scale_fill_distiller(palette = "Spectral", breaks = bks) +
  geom_text(data = diamonds, aes(x = carat, y = price, label = after_stat(count)),
          stat="binhex", bins=30, show.legend=FALSE,
          colour="black", size=2.5)

例如计数为5809和5556的分箱仍显示为蓝色。但使用geom_bin_2d时，映射是准确的：

d + geom_bin_2d(aes(fill = after_stat(count)), bins = 30) + 
  scale_fill_distiller(palette = "Spectral", breaks = bks) +
  geom_text(data = diamonds, aes(x = carat, y = price, label = after_stat(count)),
            stat="bin_2d", bins=30, show.legend=FALSE,
            colour="black", size=2.5)

问题原因 #

1. 调色板类型不匹配：Spectral是发散型调色板，但scale_fill_distiller默认使用type="seq"（顺序型）逻辑解析调色板，导致颜色映射的区间划分错误，高数值分箱无法正确对应到调色板的深色（红色）区域。
2. 分箱计算重复的潜在偏差：geom_hex和geom_text各自独立计算分箱，即使参数相同，也可能因计算细节出现微小差异，导致文本显示的count与六边形填充的count不完全匹配。

解决方法 #

方法1：正确配置发散型调色板 #

在scale_fill_distiller中指定type="div"，并设置合适的midpoint（通常取count的中位数或均值），让发散型调色板正确映射数值范围：

# 计算分箱count的中位数作为映射中点
count_mid <- median(ggplot2:::binhex(diamonds$carat, diamonds$price, bins=30)$count)

d + geom_hex(aes(fill = after_stat(count)), bins = 30, colour = "white") + 
  scale_fill_distiller(palette = "Spectral", breaks = bks, 
                       type = "div", midpoint = count_mid) +
  geom_text(stat="binhex", bins=30, 
            aes(label = after_stat(count)),
            show.legend=FALSE, colour="black", size=2.5)

方法2：改用顺序型调色板 #

如果不需要发散型配色，直接使用顺序型调色板（如"Reds"、"Blues"），无需额外配置类型：

d + geom_hex(aes(fill = after_stat(count)), bins = 30, colour = "white") + 
  scale_fill_distiller(palette = "Reds", breaks = bks) +
  geom_text(stat="binhex", bins=30, 
            aes(label = after_stat(count)),
            show.legend=FALSE, colour="black", size=2.5)

方法3：预计算分箱数据（彻底避免分箱差异） #

先统一计算六边形分箱的结果，再基于该数据绘制图形，确保填充和文本使用完全一致的count值：

# 预计算分箱数据
hex_data <- ggplot_build(d + geom_hex(aes(fill = after_stat(count)), bins=30))$data[[1]]

# 绘制图形
d + geom_hex(data = hex_data, aes(x = x, y = y, fill = count), 
             bins = 30, colour = "white") +
  scale_fill_distiller(palette = "Spectral", breaks = bks, 
                       type = "div", midpoint = median(hex_data$count)) +
  geom_text(data = hex_data, aes(x = x, y = y, label = count),
            show.legend=FALSE, colour="black", size=2.5)

内容的提问来源于stack exchange，提问作者Crops