How to build a pipeline from data.table to magrittr and back to data.table

Question

I would like to mix data.table pipelining with magrittr pipelining. I can go from data.table to %>% but I can't figure out how to get back to [][] data.table style pipelining.

Here's an example:

> tbl = data.table(grp=c(1,1,1,2,2,2,3,3,3,4,4), y=rnorm(11))
> tbl
    grp        y
 1:   1  0.08150
 2:   1  1.51330
 3:   1 -0.26154
 4:   2 -0.12746
 5:   2  0.10747
 6:   2  0.16502
 7:   3  0.54139
 8:   3 -0.04194
 9:   3  0.02373
10:   4  2.00756
11:   4  1.05523
> tbl[, .(.N, mean(y)), by=grp][order(-N)] %>% head(n=3) %>% .[, N := NULL]
   grp      V2
1:   1 0.44442
2:   2 0.04834
3:   3 0.17439
> tbl[, .(.N, mean(y)), by=grp][order(-N)] %>% head(n=3) %>% .[, N := NULL][, plot(grp, V2)]
Error in `[.data.table`(., .[, `:=`(N, NULL)], , plot(grp, V2)) : 
  'by' or 'keyby' is supplied but not j
Calls: %>% ... freduce -> withVisible -> <Anonymous> -> [ -> [.data.table
> 

How can I go back to [][] after %>% ?

I know that this particular example could be rewritten entirely with [] and no %>%, but I'm not interested in doing that every time. I'd like a way to be able to write [][] %>% [][] patterns.

Answer 1

Both the previous answers overlook your ability to specify, to some extent, precedence. You can do it by upping the precedence of the %>% part of the code by enclosing it in {}s:

x <- data.frame(a=1:5, b=6:10)
{x %>% subset(a<4) %>% data.table()}[, mean(b)]

Not pretty, but it works:

> {x %>% subset(a<4) %>% data.table()} [, mean(b)]
[1] 7

Answer 2

You can do

 `tbl %>% filter(y>0) %>% data.table()` 

to convert the pipeline result to data.table, for example, for nicely printing the results - in a data.table way. But, unfortunately, you cannot do something like

 `tbl %>% filter(y>0) %>% data.table() [, mean(y), by=group]

I wonder, if this functionality could be added to future data.table versions - maybe through new syntax (to overcome the precedence order limitation, as "[" is performed prior to "<").