efficient rename

[romainfrancois]

http://stackoverflow.com/questions/21502465/replacement-for-rename-in-dplyr

plyr::rename makes deep copies:

> dplyr:::changes( mtcars, rename(mtcars, c("disp" = "displacement")) )
Changed variables:
             old         new
mpg          0x102a35d60 0x102a4f5a0
cyl          0x102a35e90 0x102a4f6d0
hp           0x102a360f0 0x102a4f930
drat         0x102a36220 0x102a4fa60
wt           0x102a36350 0x102a4fb90
qsec         0x102a36480 0x102a4fcc0
vs           0x102a365b0 0x102a4fdf0
am           0x102a366e0 0x102a4ff20
gear         0x102a36810 0x102a50050
carb         0x102a36940 0x102a50180
disp         0x102a35fc0
displacement             0x102a4f800

Changed attributes:
             old         new
names        0x1032f02a8 0x10595c2a0
row.names    0x102a36a70 0x102a502b0
class        0x1035f7168 0x10595cfe8

And we could have something more dplyr like:

rename( mtcars, displacement = disp )

[romainfrancois]

rename_ (just naming it like this for now) uses shallow copies:

mtcars2 <- rename_( mtcars, disp2 = disp )
> dplyr:::changes(mtcars2, mtcars)
Changed variables:
      old         new
disp2 0x10323d420
disp              0x10323d420

Changed attributes:
      old         new
names 0x10aaf5200 0x104e90558

[romainfrancois]

Needs more logic to handle the grouped case, i.e handling the labels and vars attributes.

[romainfrancois]

Hmm. Did not see #192. we have too many open issues about select. We should decide what it can do. The shallow_copy internal function makes it quite trivial to implement.

[hadley]

Yes, select should do this. I'll try and implement a few this week (or at least figure out what select should do)

[romainfrancois]

When i get a chance (travelling now), i'll add support code internally that takes a data frame and a character vector of names and does the right shallow copy thing with them.

This way on the R side w just have to calculate the wanted colimns.

[hadley]

That sounds like a reasonable implementation. Can you make it take a named vector so it can simultaneously rename and select?

[hadley]

Initial implementation in 8ccdb07 - I think this will make it easy to implement select() for any backend.

[hadley]

@romainfrancois to be precise, could you please write select_impl() that is given a data frame and a named vector and produces an output without making any copies. I'll then use that in the select() methods.

[romainfrancois]

I've put in some initial code for a back end select_impl. It takes 3 parameters:

• the data frame (or grouped data frame).
• a character vector of current names of variables we want to keep
• the names we want to give them

Depending on what we want, the 2nd and 3rd might be the same vector.

The code as it is now assumes that args 2 and 3 have the same length, that strings from 2 really are in the data frame. It handles the vars and labels attributes we use for grouping.

We could reduce 2 and 3 as a named vector if you prefer

[romainfrancois]

Done. in the select branch.

> select_impl( mtcars, c( cyl2 = "cyl" ) )
                    cyl2
Mazda RX4              6
Mazda RX4 Wag          6
Datsun 710             4
Hornet 4 Drive         6
Hornet Sportabout      8
...

[hadley]

Thanks. Just pushed changes to use select_vars() throughout, and select_impl() for data frames, tbl_df() and grouped_df(). Will merge branch tomorrow once I've updated the docs.