unique.data.table select last row in place of the first

Question

calling unique on a keyed data.table you'll have unique lines per each group. In case of duplicated lines the first will be taken. When I need the take the last instead ( in general the last temporal transaction) I use .SD[.N]

library(data.table)
library(microbenchmark)

dt <- data.table(id=sample(letters, 10000, T), var=rnorm(10000), key="id")

microbenchmark(unique(dt), dt[, .SD[.N], by=id])
Unit: microseconds
                   expr      min        lq    median       uq        max neval
             unique(dt)  570.882  586.1155  595.8975  608.406   3209.122   100
 dt[, .SD[.N], by = id] 6532.739 6637.7745 6694.3820 6776.968 208264.433   100

do you know a faster way to do the same?

Answer 1

Create a data.table that contains the unique combinations of the key variables then join using mult = 'last'

Using .SD is convenient, but slow. You could use .I instead if you wished.

dtu <- unique(dt)[,key(dt), with = FALSE]
dt[dtu, mult = 'last']

Or

 dt[ dt[,  .I[.N], by = key(dt)]$V1]

Answer 2

From data.table v1.9.4, you can use fromLast = TRUE.

microbenchmark(unique(dt, by = "id"), 
               dt[, .SD[.N], by=id], 
               unique(dt, by = "id", fromLast = TRUE))

Unit: microseconds
                                   expr     min       lq     mean   median       uq      max neval cld
                  unique(dt, by = "id") 333.978 355.1900 406.1585 371.1360 393.4015 3203.769   100  a 
                 dt[, .SD[.N], by = id] 519.320 541.4345 580.2176 553.6200 563.5490 2690.167   100   b
 unique(dt, by = "id", fromLast = TRUE) 338.190 366.4725 430.1296 380.9145 400.7730 4774.663   100  a