na.strings="0" is not permitted in fread

[msgoussi]

When I use fread, I get this error (NAstring <<0>> is recognized as type boolean, this is not permitted)
na.strings = c("", "-", "_", "..", "...", "--", "**", "" ,
"n/a", "n.a.", "#VALUE!", "0", "Inf", "-Inf", "NAN", "r", "e")

If I removed "0" from na.strings, fread is not getting error.

However, cells that contains "r" or "e", is not converted to NA, and their columns are characters.

Please advise.

[MichaelChirico]

Please include your data.table package version & system info, and the file itself, if you can.

[HughParsonage]

data.table::fread("a,b,c\n0,1,2", na.strings = "0", verbose = TRUE)
#> Input contains a \n or is "". Taking this to be text input (not a filename)
#> [01] Check arguments
#>   Using 12 threads (omp_get_max_threads()=12, nth=12)
#> Error in data.table::fread("a,b,c\n0,1,2", na.strings = "0", verbose = TRUE):
#>   freadMain: NAstring <<0>> is recognized as type boolean, this is not permitted.

[msgoussi]

Sys.info()
sysname release version nodename
"Windows" ">= 8 x64" "build 9200" "IRT-310677-Z440"
machine login user effective_user
"x86-64" "310677" "310677" "310677"

data.table package: 1.11.4
The file is so huge, it is almost 1.44 GB

[MichaelChirico]

Thank @HughParsonage, darn that looks no bueno.

I guess there should be some interaction with logical01 but there doesn't appear to be.

[msgoussi]

Is it a bug?

[jangorecki]

@msgoussi yes, you can workaround it with something like

data.table::fread("a,b,c\n0,1,2")[a==0, a:=NA]

[msgoussi]

If i have 500 columns and i need to clean columns and consider the following strings , c("", "-", "_", "..", "...", "--", "**", "" ,
"n/a", "n.a.", "#VALUE!", "0", "Inf", "-Inf", "NAN", "r", "e"), as na. The way around will not look good. This is my opnion

[MichaelChirico]

you're correct, but until the bug is fixed, you can at least get rid of all the other NA strings with fread, then only replace 0 in the code.

…

On Tue, Jun 12, 2018, 5:36 AM msgoussi ***@***.***> wrote: If i have 500 columns and i need to clean colmuns and consider the following strings , c("", "-", "_", "..", "...", "--", "**", "" , "n/a", "n.a.", "#VALUE!", "0", "Inf", "-Inf", "NAN", "r", "e"), as na. The way around will not look good. This is my opnion — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2927 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHQQdRcsWHnfXUeJFvIlQ3a7WLUeMyyRks5t7uLBgaJpZM4UiP9z> .

[ribailey]

I have an R package on Github, ribailey/gghybrid, which includes a function to read genomic data files with potentially millions of columns, and one of the main softwares for producing these input files, PLINK, always codes missing data as zero. I use fread within my function to read in the data and declare missing values. This means I can't read in the most common file type people might want to use, due to the bug described here. Has there been any progress on this? Many thanks, Richard.

[ben-schwen]

Removing || strcmp(ch,"1")==0 || strcmp(ch,"0")==0 from fread.c seems like an option to "fix" this.

Only breaks 1 test case which is explicitly testing for na.strings = '1'.

Interaction with logical01 seems also legit.

data.table::fread("a,b,c\n0,1,2\n1,0,2", na.strings = "0", logical01=T)

        a      b     c                                                                                                     
    <lgcl> <lgcl> <int>                                                                                                  
1:     NA   TRUE     2                                                                                                  
2:   TRUE     NA     2 

[shrektan]

I think the expected behavior is that we should not allow logical01=TRUE and na.strings = "0" at the same time. If this is agreed, I'm happy to start to file a PR for this.

[MichaelChirico]

Agreed @shrektan. Nor na.strings = "1" though I guess that's a pretty obscure use case.