您正在寻找 split

split(df, with(df, interaction(v1,v2)), drop = TRUE)
$E.X
  v1 v2 v3 v4 v5
3  E  X  2 12 15
5  E  X  2 14 16

$D.Y
  v1 v2 v3 v4 v5
2  D  Y 10 12  8

$A.Z
  v1 v2 v3 v4 v5
1  A  Z  1 10 12

如意见中所述

以下任何一项都可以使用

any of the following would work

library(microbenchmark)
microbenchmark(
                split(df, list(df$v1,df$v2), drop = TRUE), 
               split(df, interaction(df$v1,df$v2), drop = TRUE),
               split(df, with(df, interaction(v1,v2)), drop = TRUE))


Unit: microseconds
                                                  expr      min        lq    median       uq      max neval
            split(df, list(df$v1, df$v2), drop = TRUE) 1119.845 1129.3750 1145.8815 1182.119 3910.249   100
     split(df, interaction(df$v1, df$v2), drop = TRUE)  893.749  900.5720  909.8035  936.414 3617.038   100
 split(df, with(df, interaction(v1, v2)), drop = TRUE)  895.150  902.5705  909.8505  927.128 1399.284   100

出现交互稍快一些（可能是因为 f = list 。）只是转换为函数内的交互）

It appears interaction is slightly faster (probably due the fact that the f = list(...) are just converted to an interaction within the function)

编辑

如果你只是想使用子集data.frames，那么我就会生气d建议使用data.table方便编码

If you just want use the subset data.frames then I would suggest using data.table for ease of coding

library(data.table)

dt <- data.table(df)
dt[, plot(v4, v5), by = list(v1, v2)]