unique names for data.table + lag for unbalanced panel

[matthieugomez]

• solves issue Data.table and group named "V1"  #615
• proposes solution for bug when object in the global environment is called dt or vars.

   library(dplyr)
   library(data.table)
   DT <- as.data.table(mtcars) 
   dt <- 6
   DT %>% filter(cyl==dt)

Solution is is only implemented for filter_.data.table and filter_.grouped_dt as I was not sure it was the right way to do it.

[hadley]

Oh hmmm, I think a better solution would be to use just a non-syntactic name like _DT

[matthieugomez]

Ok. I changed it.

[matthieugomez]

I've added an option along_with for the functions lead and lag, which can be used instead of order_by. While lag(var, 1, order_by = date) creates a lagged variable based on previous date, lag(var, 1, along_with = date) created a lagged variable based on date - 1.

This is an important command (at least in social sciences) and the current solution (left_join) is not as satisfying (wrt syntax and performance). Related discussions:
http://stackoverflow.com/questions/25649056/create-lagged-variable-in-unbalanced-panel-data-in-r
https://groups.google.com/forum/#!searchin/manipulatr/lag/manipulatr/xDZjSuexXE0/d5LxZTJ4DHgJ
http://r.789695.n4.nabble.com/lags-for-unbalanced-panel-data-td3341764.html
http://aidcomplexity.blogs.casa.ucl.ac.uk/2012/08/16/lagging-panel-data-in-r/

Update: n can now be a lubridate period, which gives a good syntax to work with dates

library(lubridate)
df <- data.frame(     
  id = c("1", "1", "1", "1"),
  date = mdy(c("03/01/1992", "04/03/1992", "07/15/1992", "08/21/1992")),
  value = c(4.1, 4.5, 3.3, 5.3)
)
df <- df %>% mutate(date = floor_date(date, "month"))
df %>% group_by(id) %>% mutate(lag(value, months(1), along_with = date)) 

[matthieugomez]

Inplace = TRUE creates a variable equal to true in current version of dplyr. I corrected it and added the option in select.

[matthieugomez]

added that filter() creates new data in vignette about memory

[matthieugomez]

add merge on different variable names (like data.frames) #694 . Moreover, copy data.table only if data.tables must be rekeyed or if variables must be renamed.

[hadley]

I'd prefer to wait on shallow copies here

[hadley]

I think this is a good idea in principle, but I've started adding a better abstraction - take a look at the new dt_subset()

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/