[datatable-help] constructing expressions for the j argumentfromcharacter vectors

[Matthew Dowle]

Something like this :

> DT = as.data.table(testData)
> f = function(x)length(unique(x))
> vars = "dx"
> mean(DT[,lapply(.SD,f),by="id",.SDcols=vars])[-1]
     dx
44.2212
> vars = c("dx","rx")
> mean(DT[,lapply(.SD,f),by="id",.SDcols=vars])[-1]  # same again, just 
> different vars
     dx      rx
44.2212 48.7814
> vars = c("dx","rx","clinic")
> mean(DT[,lapply(.SD,f),by="id",.SDcols=vars])[-1]  # same again, just 
> different vars
     dx      rx  clinic
44.2212 48.7814  9.9331
>

Chris' suggestion of parse(text=paste(...)) is another way you could do it 
(and may be more efficient).

Matthew


"Matthew Dowle" <mdowle at mdowle.plus.com> wrote in message 
news:j5vdoq$epb$1 at dough.gmane.org...
> Hi,
>
> Welcome.
> Just to check you've found .SD,  [,lapply(.SD,sum),by=...], and .SDcols?
> .SD consist of all columns other than the grouping columns, which seems 
> similar
> to what this line is doing? :
>> mean(summaryDT[,(ncols-length(sList) + 2):ncols, with = FALSE])
>
> Matthew
>
>
> "Erik Iverson" <erikriverson at gmail.com> wrote in message 
> news:CAKzGw12zWpPt3pSqJCDH_SmDOQOLAjRUV7cV64UYWb8pXK13uQ at mail.gmail.com...
> Hello,
>
> Thank you for providing the data.table package, I think it will be
> very useful to me going forward.  I have a question about passing
> around expressions, and have come up with an example to show what I'm
> after.
>
> library(data.table)
> ## test data
> N <- 500000
> set.seed(100)
> testData <- data.frame(id = c(sample(1:10000, N, replace = TRUE)),
> clinic = c(sample(1:10, N, replace = TRUE)),
> dx = c(sample(1:200, N, replace = TRUE)),
> rx = c(sample(1:1000, N, replace = TRUE)))
>
> ## want to know mean number of dx per ID
> mean(tapply(testData$dx, testData$id,
> function(x) length(unique(x)))) ## 44.2212
>
> ## in my real use case, I want to run this with different 'by'
> ## variables, so let's write a function and try to use data.table,
> ## call the function uniqueSummary1
>
> uniqueSummary1 <- function(df, key) {
> DT <- data.table(df)
> key(DT) <- key
>
> summaryDT <- DT[, list(length(unique(dx)),
> length(unique(rx))), by = key]
>
> mean(summaryDT[,list(V1, V2)])
>
> }
>
> ## agrees with tapply
> uniqueSummary1(df = testData, key = c("id"))
>
> ## The above works great, but isn't general, since in my real use
> ## case, I won't know dx and rx are the variables of interest. I want
> ## to be able to pass them in as arguments. This is exactly what FAQ
> ## 1.6 is, so let's use that solution to define uniqueSummary2
>
> uniqueSummary2 <- function(df, key, vars) {
> DT <- data.table(df)
> key(DT) <- key
>
> sList <- substitute(vars)
> summaryDT <- DT[, eval(sList), by = key]
> ncols <- ncol(summaryDT)
>
> mean(summaryDT[,(ncols-length(sList) + 2):ncols, with = FALSE])
> }
>
> uniqueSummary2(df = testData, key = c("id"),
> vars = list(length(unique(dx)),
> length(unique(rx)),
> length(unique(clinic))))
>
> ## uniqueSummary2 is better, but relies on me repeating the
> ## "length(unique())" bit several times. Ideally, I'd just like to
> ## pass in a list of QUOTED vars to summarize, like the following
> ## hypothetical call to my yet-unwritten uniqueSummary3 function:
>
> uniqueSummary3(df = testData, key = c("id"),
> vars = c("dx", "rx", "clinic"))
>
> I assume I can somehow construct the expression for the j index inside
> my function, based on the 'vars' character vector, but am stuck on
> how.  Any ideas?
>
> Thanks so much,
> Erik