[datatable-help] number of rows selected in .SD subset

[Ben Tupper]

Hi Arun,

The vignette is very helpful; the ?data.table help page is so rich and dense that I end up wandering quite a bit.  The vignette does a nice job laying it out logically.  I'm sure it has been a huge effort.

> DT[, .SD[.N], by=month] # ~ since .N contains the number of observations in this group


Doh! Now I see it is stated clearly right under my nose: ".N is a special in-built variable that holds the number of observations in the current group."  I'm not sure why I thought .N was for the original data.table instead of the grouping.

I have switched my script to use the above and it is lightning fast now.  I'm going to start wearing a seatbelt and helmet...

Thank you!
Ben

P.S.  How did you get download.file to read an https URL for the example?  Here's what I get...
flights <- fread("https://raw.githubusercontent.com/wiki/arunsrinivasan/flights/NYCflights14/flights14.csv")
Error in download.file(input, tt, mode = "wb") : unsupported URL scheme
So, I downloaded it manually using a browser and used my local copy instead.




On Jan 22, 2015, at 2:11 PM, Arunkumar Srinivasan <aragorn168b at gmail.com> wrote:

> Ben,
> 
> Great to hear that you're going thro' the vignette..
> 
> To get the last row, you can similarly do:
> 
> DT[, tail(.SD, 1L), by=month] # ~ as you say
> DT[, .SD[.N], by=month] # ~ since .N contains the number of observations in this group
> DT[, .SD[(.N-1L):.N], by=month] # ~ last two rows per group
> 
> However, `.SD[...]` per group is slightly slower (especially on many groups) as it has to go through `[.data.table` (which is a S3 generic, and takes time for dispatching the right method.. which can get noticeable on large groups), and not all cases are optimised. 
> 
> You can also use `.I` (which is deliberately not mentioned in the vignette to keep things smooth and straightforward). Using it you could do:
> 
> idx = DT[, .I[1L], by=month][, V1]
> DT[idx]
> 
> `.I` contains the row number in `x` (it doesn't reset per group..). So we can get the row indices for each group for the first element, and then simply subset. We hope to improve this subset in the future (to take care of this optimisation internally).
> 
> Similarly:
> 
> idx = DT[, .I[.N], by=month][, V1]
> DT[idx]
> 
> will get the last element for each group.
> 
> Otherwise, how do you find the vignette so far?
> 
> HTH,
> Arun
> 
> 
> 
> On Thu, Jan 22, 2015 at 6:11 PM, Ben Tupper <btupper at bigelow.org> wrote:
> Hello,
> 
> I have been learning to use data.table and studying the vignette located here...
> 
> https://rawgit.com/wiki/Rdatatable/data.table/vignettes/datatable-intro-vignette.html
> 
> Section 2f. shows how to subset a data.table to select an arbitrary number of rows in each .SD.  That's really handy.
> 
> 2. Aggregations
>   f. Subset .SD for each group:    ans <- flights[, head(.SD, 2), by=month]
> 
> In a similar way, I can get the last row of the .SD using either tail, nrow or dim (I don't think it matters much, but dim seems to be a faster*).
> 
>   ans <- flights[,.SD[dim(.SD)[1]], by=month]
> 
> I got to wondering if the number of rows in .SD might be exposed in each grouping iteration.  Is there an equivalent to .N for the subset data.table, .SD?  Something like .SDN or the like?
> 
> Thanks for data.table!
> 
> Ben
> 
> * After reading this discussion http://r.789695.n4.nabble.com/What-is-the-fastest-way-to-determine-that-data-table-is-empty-td4638348.html#a4638451 I tried out a couple of methods for getting the last element of a grouping using nrow(), tail() and dim().
> 
> # using tail
> > microbenchmark( last1 <- flights[, tail(.SD, 1), by=month] )
> Unit: milliseconds
>                                          expr      min       lq     mean   median       uq      max neval
>  last1 <- flights[, tail(.SD, 1), by = month] 16.65898 16.89704 18.26415 17.37007 19.20147 40.12966   100
> 
> # using dim
> >   microbenchmark( last2 <- flights[,.SD[dim(.SD)[1]], by=month] )
> Unit: milliseconds
>                                              expr      min       lq     mean   median       uq      max neval
>  last2 <- flights[, .SD[dim(.SD)[1]], by = month] 15.51243 15.87788 17.40978 16.19426 17.83308 59.22429   100
> 
> # using nrow
> >   microbenchmark( last3 <- flights[,.SD[nrow(.SD)], by=month] )
> Unit: milliseconds
>                                            expr      min       lq     mean   median       uq      max neval
>  last3 <- flights[, .SD[nrow(.SD)], by = month] 15.63919 15.92073 17.28836 16.52588 18.33867 24.92624   100
> 
> >   identical(last1, last2)
> [1] TRUE
> >   identical(last1, last3)
> [1] TRUE
> 
> Ben Tupper
> Bigelow Laboratory for Ocean Sciences
> 60 Bigelow Drive, P.O. Box 380
> East Boothbay, Maine 04544
> http://www.bigelow.org
> 
> 
> 
> 
> 
> 
> 
> 
> _______________________________________________
> 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
> 

Ben Tupper
Bigelow Laboratory for Ocean Sciences
60 Bigelow Drive, P.O. Box 380
East Boothbay, Maine 04544
http://www.bigelow.org








-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/datatable-help/attachments/20150122/51a0e28f/attachment-0001.html>