数据表到嵌套列表

我想转换：

library(data.table)    
n <- 12
DT <- data.table(
  level1 = rep(paste0("Manu", 1:2), each = n / 2),
  level2 = rep(paste0("Dept", 1:4), each = n / 4),
  level3 = rep(paste0("Store", 1:n))
)
> DT
level1 level2  level3
1:  Manu1  Dept1  Store1
2:  Manu1  Dept1  Store2
3:  Manu1  Dept1  Store3
4:  Manu1  Dept2  Store4
5:  Manu1  Dept2  Store5
6:  Manu1  Dept2  Store6
7:  Manu2  Dept3  Store7
8:  Manu2  Dept3  Store8
9:  Manu2  Dept3  Store9
10:  Manu2  Dept4 Store10
11:  Manu2  Dept4 Store11
12:  Manu2  Dept4 Store12

To this:

goal <- list(
  Manu1 = list(
    Dept1 = paste0("Store", 1:(n / 4)),
    Dept2 = paste0("Store", (n/4 + 1):(n / 2))
  ),
  Manu2 = list(
    Dept3 = paste0("Store", (n/2 + 1):(3 * n / 4)),
    Dept4 = paste0("Store", (3 * n / 4 + 1):n)
  )
)
> goal
$Manu1
$Manu1$Dept1
[1] "Store1" "Store2" "Store3"

$Manu1$Dept2
[1] "Store4" "Store5" "Store6"


$Manu2
$Manu2$Dept3
[1] "Store7" "Store8" "Store9"

$Manu2$Dept4
[1] "Store10" "Store11" "Store12"

什么是data.table方法来做到这一点？

借用 @eddi 的评论（需要更新data.table至 1.9.8+）：

s = split(DT, by = c('level1', 'level2'), keep.by = FALSE, flatten = FALSE)
rapply(relist(DT[['level3']], s), unname, how="replace")

$Manu1
$Manu1$Dept1
[1] "Store1" "Store2" "Store3"

$Manu1$Dept2
[1] "Store4" "Store5" "Store6"


$Manu2
$Manu2$Dept3
[1] "Store7" "Store8" "Store9"

$Manu2$Dept4
[1] "Store10" "Store11" "Store12"

从计算角度来看，这看起来相当浪费（在树结构上迭代三次），但至少它应该扩展到比两层更深的嵌套（感谢split.data.table在 1.9.8+ 中）。