Empty column names need to be given unique names

[jabranham]

I used readr::read_csv() to import a csv file into R, which resulted in no errors. Later, when trying to run lm() on the data i got the error

Error in terms.formula(formula, data = data) : 
  attempt to use zero-length variable name

I found this question/answer on stackexchange, which seems to indicate that other people are having the same issue.

http://stackoverflow.com/questions/31385976/error-attempt-to-use-zero-length-variable-name

[hadley]

Could you please provide a minimal reproducible example (preferably using a small made up dataset)?

[jabranham]

Yes, this is so simple that I feel like I'm just making some sort of silly mistake. But here is code that replicates it for me.

set.seed(2904)

thedata <- data.frame(
  x = rnorm(100),
  y = rnorm(100, 3, 270),
  groups = rep(1:5))

write.csv(thedata, "examplecsv.csv")

thedata2 <- readr::read_csv("examplecsv.csv")

lm(y ~ x * groups, data = thedata2)

And the results of sessionInfo, if you want it:

R version 3.2.3 (2015-12-10)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Arch Linux

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

loaded via a namespace (and not attached):
[1] compiler_3.2.3 readr_0.2.2    tools_3.2.3    Rcpp_0.12.3   

[jennybc]

The write.csv() command puts row names into the file and readr::read_csv() brings them back in as a variable with the empty string as a name. Which apparently lm() doesn't like, even though the model has nothing to do with that variable.

You could prevent the writing of row names via write.csv(thedata, "examplecsv.csv", row.names = FALSE) or prevent the reading of them via readr::read_csv("examplecsv.csv", col_types = "_nni"). Or drop them once you read the data in, thedata2[names(thedata2) != ""] but before you fit the model.

Which, I now see, is pretty much what the SO answers say.

Look for a variable named "" in your data and either drop it or rename it.

UPDATE: Hey, I recognize you from Austin!

[jabranham]

Ah, that makes sense. Might it make sense for read_csv to name the rownames something if they aren't named (like row_names)? When I use utils::read.csv() it names them X.

And why in the world does lm complain about variables that aren't in the model? That's just weird.

Yup, I was at your talk in Austin. I've taught a few people git and show them your "burn it all down" slides.

Update: This is somewhat related to tidyverse/dplyr#1576

[jennybc]

I think @hadley might refine this particular point of the readr philosophy: "Column names are left as is". Maybe an exception will be made to populate missing names? And maybe the empty-string-as-variable-name would get the same treatment.

[hadley]

Yes, this will definitely get fixed - missing/empty col names need some repair because they cause so many downstream problems, and it's never helpful to maintain the missingness.

Will probably adopt some convention like _missing_1, _missing_2 for missing names.

[hadley]

We now have:

read_csv(",,\n1,2,3")
#> Source: local data frame [1 x 3]
#> 
#>      X1    X2    X3
#>   <int> <int> <int>
#> 1     1     2     3

But this probably needs an explicit warning

[hadley]

It doesn't seem useful to generate an invalid data frame, so now both missing and duplicated column names get an automatic fix and a warning:

x1 <- read_csv(",,\n1,2,3")
#> Warning: Missing column names filled in: 'X1' [1], 'X2' [2], 'X3' [3]
x2 <- read_csv("x,x,x\n1,2,3")
#> Warning: Duplicated column names deduplicated: 'x' => 'x_1' [2], 'x' =>
#> 'x_2' [3]
x3 <- read_csv("X2,\n1,2")
#> Warning: Missing column names filled in: 'X2' [2]
#> Warning: Duplicated column names deduplicated: 'X2' => 'X2_1' [2]

[cluoma]

Is it possible to disable 'deduplicating' column names? This is undesired behaviour in my case since the csv is malformed such that the first and second rows together create unique column names.

[hadley]

Just use col_names = FALSE

[mightypog]

I did a completely low tech thing. I pulled the empty column out of the original spreadsheet and reloaded it. Not a coding solution, but it worked.