[R] lapply or data.table to find a unit's previous transaction

Matthew Dowle mdowle at mdowle.plus.com
Thu Jun 3 12:10:07 CEST 2010


William,

Try a rolling join in data.table, something like this (untested) :

setkey(Data, UnitID, TranDt)    # sort by unit then date
previous = transform(Data, TranDt=TranDt-1)
Data[previous,roll=TRUE]        # lookup the prevailing date before, if any, 
for each row within that row's UnitID

Thats all it is, no loops required. That should be fast and memory 
efficient. 100's of times faster than a subquery in SQL.

If you have trouble please follow up on datatable-help.

Matthew


"William Rogers" <whrogers73 at gmail.com> wrote in message 
news:AANLkTikk_aVUPm7J108isEryO9FuCpnJHaNXPAqVT7_o at mail.gmail.com...
I have a dataset of property transactions that includes the
transaction ID (TranID), property ID (UnitID), and transaction date
(TranDt). I need to create a data frame (or data table) that includes
the previous transaction date, if one exists.
This is an easy problem in SQL, where I just run a sub-query, but I'm
trying to make R my one-stop-shopping program. The following code
works on a subset of my data, but I can't run this on my full dataset
because my computer runs out of memory after about 30 minutes. (Using
a 32-bit machine.)
Use the following synthetic data for example.

n<- 100
TranID<- lapply(n:(2*n), function(x) (
as.matrix(paste(x, sample(seq(as.Date('2000-01-01'),
as.Date('2010-01-01'), "days"), sample(1:5, 1)), sep= "D"), ncol= 1)))
TranID<- do.call("rbind", TranID)
UnitID<- substr(TranID, 1, nchar(n))
TranDt<- substr(TranID, nchar(n)+2, nchar(n)+11)
Data<- data.frame(TranID= TranID, UnitID= UnitID, TranDt= as.Date(TranDt))

#First I create a list of all the previous transactions by unit

TranList<- as.matrix(Data$TranID, ncol= 1)
PreTran<- lapply(TranList,
function(x) (with(Data,
Data[
UnitID== substr(x, 1, nchar(n))&
TranDt< Data[TranID== x, "TranDt"], ]
))
)

#I do get warnings about missing data because some transactions have
no predecessor.
#Some transactions have no previous transactions, others have many so
I pick the most recent

BeforeTran<- lapply(seq_along(PreTran), function(x) (
with(PreTran[[x]], PreTran[[x]][which(TranDt== max(TranDt)), ])))

#I need to add the current transaction's TranID to the list so I can merge 
later

BeforeTran<- lapply(seq_along(PreTran), function(x) (
transform(BeforeTran[[x]], TranID= TranList[x, 1])))

#Finally, I convert from a list to a data frame

BeforeTran<- do.call("rbind", BeforeTran)

#I have used a combination of data.table and for loops, but that seems
cheesey and doesn't preform much better.

library(data.table)

#First I create a list of all the previous transactions by unit

TranList2<- vector(nrow(Data), mode= "list")
names(TranList2)<- levels(Data$TranID)
DataDT<- data.table(Data)

#Use a for loop and data.table to find the date of the previous transaction

for (i in levels(Data$TranID)) {
if (DataDT[UnitID== substr(i, 1, nchar(n))&
TranDt<= (DataDT[TranID== i, TranDt]),
length(TranDt)]> 1)
TranList2[[i]]<- cbind(TranID= i,
DataDT[UnitID== substr(i, 1, nchar(n))&
TranDt< (DataDT[TranID== i, TranDt]),
list(TranDt= max(TranDt))])
}

#Finally, I convert from a list to a data table

BeforeTran2<- do.call("rbind", TranList2)

#My intution says that this code doesn't take advantage of
data.table's attributes.
#Are there any ideas out there? Thank you.
#P.S. I've tried plyr and it does not help my memory problem.

--
William H. Rogers



More information about the R-help mailing list