as_tibble.matrix should keep rownames attribute

[ClaytonJY]

Currently, as_tibble.matrix() throws away rownames:

library(tibble)

mtmatrix <- as.matrix(mtcars)
rownames(mtmatrix)
#>  [1] "Mazda RX4"           "Mazda RX4 Wag"       "Datsun 710"         
#>  [4] "Hornet 4 Drive"      "Hornet Sportabout"   "Valiant"            
#>  [7] "Duster 360"          "Merc 240D"           "Merc 230"           
#> [10] "Merc 280"            "Merc 280C"           "Merc 450SE"         
#> [13] "Merc 450SL"          "Merc 450SLC"         "Cadillac Fleetwood" 
#> [16] "Lincoln Continental" "Chrysler Imperial"   "Fiat 128"           
#> [19] "Honda Civic"         "Toyota Corolla"      "Toyota Corona"      
#> [22] "Dodge Challenger"    "AMC Javelin"         "Camaro Z28"         
#> [25] "Pontiac Firebird"    "Fiat X1-9"           "Porsche 914-2"      
#> [28] "Lotus Europa"        "Ford Pantera L"      "Ferrari Dino"       
#> [31] "Maserati Bora"       "Volvo 142E"
rownames(as_tibble(mtmatrix))
#>  [1] "1"  "2"  "3"  "4"  "5"  "6"  "7"  "8"  "9"  "10" "11" "12" "13" "14"
#> [15] "15" "16" "17" "18" "19" "20" "21" "22" "23" "24" "25" "26" "27" "28"
#> [29] "29" "30" "31" "32"

This is inconsistent with as_tibble.data.frame, which makes sure to keep rownames

all.equal(
  rownames(mtcars),
  rownames(as_tibble(mtcars))
)
#> [1] TRUE

If a user has a rownamed matrix and wants a tibble with those rownames in it (e.g. confusion matrices from the randomForest package), they have to convert to a base data.frame first

mtmatrix %>%
  as.data.frame() %>%
  as_tibble() %>%
  rownames_to_column("rname")

Fixing this bug would remove the need for the as.data.frame() call, and make the steps match those for doing the same with a data.frame, e.g.

mtcars %>%
  as_tibble() %>%
  rownames_to_column("rname")

This only saves one step, but standardizing the tibble interface across matrices and data.frames where possible seems like a worthy UX improvement.

Session Info

Used newest github version to make sure this wasn't fixed already

devtools::session_info("tibble")
#> Session info --------------------------------------------------------------
#>  setting  value                       
#>  version  R version 3.4.1 (2017-06-30)
#>  system   x86_64, linux-gnu           
#>  ui       X11                         
#>  language (EN)                        
#>  collate  en_US.UTF-8                 
#>  tz       America/New_York            
#>  date     2017-08-02
#> Packages ------------------------------------------------------------------
#>  package * version    date       source                           
#>  Rcpp      0.12.12    2017-07-15 cran (@0.12.12)                  
#>  rlang     0.1.1      2017-05-18 CRAN (R 3.4.0)                   
#>  tibble  * 1.3.3.9001 2017-08-02 Github (tidyverse/tibble@12767ae)

[ClaytonJY]

Looks like this conversion is done in C-land with matrixToDataFrame; does the fix need to happen in there?

That's a bit outside my skills, but I can try re-setting the attribute after the conversion, inside as_tibble.matrix, and put that in a PR with a test or two.

[krlmlr]

Thanks. I agree that there is some inconsistency, but I'd rather not support row names in tibbles altogether. We may want to give a warning at some point, so there seems to be no point in enabling this functionality now.

We may want to consider adding a key argument instead, once we figure out support for keys (tidyverse/dplyr#1792), a special value would then convert the row names to a column and also work for matrices.

[ClaytonJY]

@krlmlr I think the key idea is really interesting, and something I've definitely missed from data.table, so it's exciting to see work in that direction.

I agree that tibble shouldn't support rownames, but I think it's important to make it easy for users to work with data that has rownames through no fault of theirs. That's why I made the PR, to make it as easy to get matrix rownames into a column as it is for data.frames.

Even better might be to have the as_tibble methods for matrix/data.frames accept a "rownames" argument, which will do what rownames_to_column does up-front, thus avoiding the need to keep the rownames attribute on the tibble at all. Is that what you're thinking the keys will allow for?

[ClaytonJY]

If that wouldn't be handled best with new keys functionality, I'd be happy to write up a PR to modify as_tibble.matrix and as_tibble.data.frame to do the rownames-to-column conversion up front!

[krlmlr]

as_tibble(rownames = "id") works now for matrices and data frames.