[datatable-help] behavior of unique on data.tables with strings

Steven C. Bagley steven.bagley at gmail.com
Tue Jan 3 05:13:30 CET 2012


It still happens. (I deleted R and all packages, then reinstalled just to check.) 

test.data.table() completes without errors. 

Here's the session info.

> sessionInfo()
R version 2.14.0 (2011-10-31)
Platform: x86_64-apple-darwin9.8.0/x86_64 (64-bit)

locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

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

other attached packages:
[1] data.table_1.7.7

> .Machine$double.eps ^ 0.5
[1] 1.490116e-08

--Steve

On Jan 2, 2012, at 3:27 PM, Matthew Dowle wrote:

> Thanks for the nice report. Oddly though, it seems to work ok for me
> both in 1.7.7 and latest 1.7.8.
> 
> $ R --vanilla
> R version 2.14.1 (2011-12-22)
> Platform: i686-pc-linux-gnu (32-bit)
>> require(data.table)
> Loading required package: data.table
> data.table 1.7.7  For help type: help("data.table")
>> foo2=as.data.table(data.frame(a=c("1", "1"), b=c(2,2),
> stringsAsFactors=FALSE)) 
>> unique(foo2)
>     a b
> [1,] 1 2
>> str(foo2)
> Classes ‘data.table’ and 'data.frame':	2 obs. of  2 variables:
> $ a: chr  "1" "1"
> $ b: num  2 2
>> .Machine$double.eps ^ 0.5
> [1] 1.490116e-08
> 
> Could you rerun and confirm please. If you are 64bit, please include
> sessionInfo(). I've included tolerance as a long shot - the numeric 2's
> are considered equal by data.table's unique() using tolerance. Perhaps
> that part is not working for you. Does test.data.table() work? It should
> test unique and tolerance fairly thoroughly. Otherwise I can't think why
> the character column isn't liked by unique, should be ok.
> 
> A fast unique for character columns is a good feature request, please
> could you add to the tracker. That is now possible to implement as we
> now have fast character methods.
> 
> Matthew
> 
> On Mon, 2011-12-26 at 19:33 -0800, Steven C. Bagley wrote:
>> In data.table 1.7.7: 
>> 
>> The function unique works for datatables (without keys) that have factors, but not if they have strings. In the latter case, setting the key will convert the strings to factors. I can't figure out from the documentation if this is the intended behavior or not. (The documentation does say that keys can't be characters/strings). It would be nice if unique would work without having to convert strings to factors because of the conversion cost in very large datatables, but maybe this isn't possible.
>> 
>> --Steve
>> 
>>> library(data.table)
>>> foo1=as.data.table(data.frame(a=c("1", "1"), b=c(2,2)))
>>> foo1
>>     a b
>> [1,] 1 2
>> [2,] 1 2
>>> str(foo1)
>> Classes ‘data.table’ and 'data.frame':	2 obs. of  2 variables:
>> $ a: Factor w/ 1 level "1": 1 1
>> $ b: num  2 2
>>> unique(foo1)
>>     a b
>> [1,] 1 2
>>> foo2=as.data.table(data.frame(a=c("1", "1"), b=c(2,2), stringsAsFactors=FALSE))
>>> foo2
>>     a b
>> [1,] 1 2
>> [2,] 1 2
>>> str(foo2)
>> Classes ‘data.table’ and 'data.frame':	2 obs. of  2 variables:
>> $ a: chr  "1" "1"
>> $ b: num  2 2
>>> unique(foo2)
>>     a b
>> [1,] 1 2
>> [2,] 1 2
>>> setkey(foo2, a)
>>> str(foo2)
>> Classes ‘data.table’ and 'data.frame':	2 obs. of  2 variables:
>> $ a: Factor w/ 1 level "1": 1 1
>> $ b: num  2 2
>> - attr(*, "sorted")= chr "a"
>>> unique(foo2)
>>     a b
>> [1,] 1 2
>> _______________________________________________
>> datatable-help mailing list
>> datatable-help at lists.r-forge.r-project.org
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/datatable-help
> 
> 



More information about the datatable-help mailing list