使用setdiff的r data.table聚合函数

Question

我有一个包含三列的data.table，需要进行复杂的聚合：

> c
                sample     group  symbol
   1:        APPL/S Up  CBEbrown   Icosl
   2:        APPL/S Up  CBEbrown   Ampd3
   3:        APPL/S Up  CBEbrown   Thbs2
   4:        APPL/S Up  CBEbrown  Map4k4
   5:        APPL/S Up  CBEbrown Slc45a3
  ---
1724: APPL/S_BD10-2 Up TCXyellow   Nfxl1
1725: APPL/S_BD10-2 Up TCXyellow    Rhog
1726: APPL/S_BD10-2 Up TCXyellow   Wipf1
1727: APPL/S_BD10-2 Up TCXyellow Selenos
1728: APPL/S_BD10-2 Up TCXyellow  Kdelr2

因此，sample有两个结果，每个结果有7个groups。基本上，需要"APPL/S _ and 10-2 Up“中的fsetdiff of symbols，而不是”APPL/S Up“中的symbols：

setdiff(c[group == "TCXyellow" & sample == "APPL/S_BD10-2 Up", symbol], 
        c[group == "TCXyellow" & sample == "APPL/S Up", symbol])

但是我想计算每个符号，fsetdiff发生了多少个fsetdiff(从0到7可能)。输出结果如下：

> out = c[, N_diff := fsetdiff(?????), by="symbol"]
> out
      symbol  N_diff
  1:   Icosl       4
  2:   Ampd3       5
  3:   Thbs2       7
  4:  Map4k4       4
  5: Slc45a3       4
 ---
503:  Unc13d       1
504:   Rpl30       1
505:    Tpt1       1
506:  Garre1       1
507: Selenos       1

7组中有4组Icosl处于"APPL/S _ in 10-2 Up“，而不是”APPL/S Up“。

Answer 1

我想你可以这样做：

f <- function(sam,sym) setdiff(sym[sam!="APPL/S UP"], sym[sam=="APPL/S UP"])

df[,.(symbol = f(sample,symbol)),group][, .(N_diff = uniqueN(group)),symbol]