Crosstalk分组串扰问题：如何实现动态过滤数据的均值计算？

问题原因 #

你用plot_ly()管道里的group_by()+mutate()计算均值时，plotly的group_by()是用来控制绘图的trace分组，而非数据聚合。结合SharedData时，原始数据的行级信息被保留，导致mean(value)计算的是每行自身的均值（等于value本身），而非time+group分组的均值。

无需Shiny的解决方案 #

方案1：用plotly内置的transforms实现动态聚合

直接在add_trace中指定聚合规则，让plotly自动对过滤后的数据按time和group计算均值，代码更简洁：

library(crosstalk)
library(tidyverse)
library(plotly)

set.seed(12345)
d <- data.frame(
  time = rep(0:9, 10),
  group = sample(paste0("Group_", seq(5)), size = 1000, replace = TRUE),
  value = rnorm(n = 1000) + 1
)

sd1 <- SharedData$new(d )

# 使用transforms实现分组均值聚合
fig <- plot_ly(sd1) %>%
  add_trace(
    x = ~time,
    y = ~value,
    type = 'scatter',
    mode = 'lines+markers',
    color = ~group,
    transforms = list(
      list(
        type = 'aggregate',
        # 按time和group的组合分组
        groups = ~interaction(time, group),
        aggregations = list(
          list(
            target = 'y',
            func = 'avg', # 指定聚合函数为均值
            enabled = TRUE
          )
        )
      )
    )
  )

bscols(
  widths = c(3,9),
  list(
    filter_slider("f1","value",sd1, ~value),
    filter_select("f2","time",sd1, ~time)
  ),
  fig
)

方案2：创建汇总后的SharedData

利用crosstalk::compute()基于原始SharedData动态计算过滤后的分组均值，适合需要复用汇总数据的场景：

library(crosstalk)
library(tidyverse)
library(plotly)

set.seed(12345)
d <- data.frame(
  time = rep(0:9, 10),
  group = sample(paste0("Group_", seq(5)), size = 1000, replace = TRUE),
  value = rnorm(n = 1000) + 1
)

sd1 <- SharedData$new(d )

# 创建汇总后的SharedData，按time和group计算均值
sd_summary <- SharedData$new(
  d,
  key = ~interaction(time, group), # 用组合键标识分组
  group = sd1$groupName() # 和原始SharedData同组，同步过滤
) %>%
  compute(
    mean_value = mean(value),
    group_by = c("time", "group") # 指定分组字段
  )

# 用汇总数据绘图
fig <- plot_ly(sd_summary) %>%
  add_trace(
    x = ~time,
    y = ~mean_value,
    type = 'scatter',
    mode = 'lines+markers',
    color = ~group
  )

bscols(
  widths = c(3,9),
  list(
    filter_slider("f1","value",sd1, ~value),
    filter_select("f2","time",sd1, ~time)
  ),
  fig
)

说明 #

两种方案都能响应crosstalk的动态过滤，实时更新分组均值：

• 方案1无需额外处理数据，直接在plotly中配置聚合逻辑，适合快速实现需求；
• 方案2生成了独立的汇总数据对象，后续如果需要对汇总结果做更多操作（比如添加其他统计量）会更灵活。

内容的提问来源于stack exchange，提问作者Erin