如何修改ggplot2散点代码按颜色排序或让ggpointgrid兼容ggplotly？

一、修改社区代码实现按颜色排序散点 #

核心思路是手动计算每个分类单元格内的点坐标，替代geom_pointrect的布局逻辑，用原生geom_point绘图以兼容ggplotly，同时在计算坐标时按颜色分组排序，让同颜色点集中排列。

代码示例 #

library(ggplot2)
library(plotly)
library(dplyr)

# 构造示例数据
set.seed(123)
df <- tibble(
  x = sample(c("A", "B", "C"), 100, replace = TRUE),
  y = sample(c("X", "Y", "Z"), 100, replace = TRUE),
  color = sample(c("Group1", "Group2"), 100, replace = TRUE)
)

# 计算按颜色排序后的点坐标
df_pos <- df %>%
  group_by(x, y) %>%
  arrange(color, .by_group = TRUE) %>% # 按颜色排序，同颜色点集中
  mutate(
    n = n(),
    cols = ceiling(sqrt(n)), # 单元格内的列数
    rows = ceiling(n / cols), # 单元格内的行数
    idx = row_number(),
    # 计算x/y方向的偏移量，均匀分布点
    x_jitter = as.integer(factor(x)) + ((idx - 1) %% cols)/(cols + 1) - 0.5/(cols + 1),
    y_jitter = as.integer(factor(y)) + ((idx - 1) %/% cols)/(rows + 1) - 0.5/(rows + 1)
  ) %>%
  ungroup()

# 绘图并转换为ggplotly
p <- ggplot(df_pos, aes(x = x_jitter, y = y_jitter, color = color)) +
  # 绘制分类网格线，模拟geom_pointrect的背景效果
  geom_hline(aes(yintercept = as.integer(factor(y)) + 0.5), color = "gray80") +
  geom_hline(aes(yintercept = as.integer(factor(y)) - 0.5), color = "gray80") +
  geom_vline(aes(xintercept = as.integer(factor(x)) + 0.5), color = "gray80") +
  geom_vline(aes(xintercept = as.integer(factor(x)) - 0.5), color = "gray80") +
  geom_point(size = 2) +
  # 还原原始分类坐标轴标签
  scale_x_continuous(breaks = 1:nlevels(factor(df$x)), labels = levels(factor(df$x))) +
  scale_y_continuous(breaks = 1:nlevels(factor(df$y)), labels = levels(factor(df$y))) +
  theme_minimal()

ggplotly(p)

关键说明 #

• 通过arrange(color, .by_group = TRUE)在每个x-y分类单元格内按颜色排序，保证同颜色点连续排列
• 手动计算的x_jitter/y_jitter让点在单元格内均匀分布，避免重叠，效果接近geom_pointrect

二、让ggpointgrid兼容ggplotly的思路 #

ggplotly不支持自定义geom的本质是无法识别其图层计算逻辑，可通过以下两种方式解决：

方式1：提取geom_pointrect的计算坐标直接构建plotly图形 #

library(ggpointgrid)
library(plotly)

# 用ggpointgrid生成基础图
gg <- ggplot(df, aes(x, y, color = color)) + geom_pointrect(size = 2)
# 提取图层计算后的点数据
gb <- ggplot_build(gg)
point_data <- gb$data[[1]]

# 用plotly重新构建图形
p <- plot_ly(point_data, x = ~x, y = ~y, color = ~colour, type = "scatter", mode = "markers") %>%
  layout(
    xaxis = list(tickmode = "array", tickvals = unique(point_data$x), ticktext = levels(factor(df$x))),
    yaxis = list(tickmode = "array", tickvals = unique(point_data$y), ticktext = levels(factor(df$y)))
  )
p

方式2：注册自定义geom的ggplotly转换方法 #

通过给geom_pointrect添加plotly转换函数，让ggplotly能直接识别并转换该图层：

# 注册转换方法
ggplotly.geom_pointrect <- function(data, params, p, ...) {
  trace <- list(
    x = data$x,
    y = data$y,
    type = "scatter",
    mode = "markers",
    marker = list(size = params$size %||% 2, color = data$colour),
    text = data$label %||% ""
  )
  return(list(trace))
}

# 直接用ggplotly转换ggpointgrid的图
gg <- ggplot(df, aes(x, y, color = color)) + geom_pointrect()
ggplotly(gg)

内容的提问来源于stack exchange，提问作者aaa