tl;dr: What are potential problems after a truelength over-allocation warning?
Recently I did something stupid like this:
m <- matrix(seq_len(1e4),nrow=10)
library(data.table)
DT <- data.table(id=rep(1:2,each=5),m)
DT[,id2:=id]
#Warning message:
# In `[.data.table`(DT, , `:=`(id2, id)) :
# tl (2002) is greater than 1000 items over-allocated (ncol = 1001).
# If you didn't set the datatable.alloccol option very large,
# please report this to datatable-help including the result of sessionInfo().
DT[,lapply(.SD,mean),by=id2]
After some searching it became apparent that the warning resulted from adding a column by reference to a data.table with too many columns and I found some rather technical explanations (e.g., this), which I probably don't fully understand.
I know that I can avoid the issue (e.g., use data.table(id=rep(1:2,each=5),stack(as.data.frame(m)))
), but I wonder if I should expect problems subsequent to such a warning (other than the obvious performance disadvantage from working with a wide format data.table).
R version 2.15.3 (2013-03-01)
Platform: x86_64-w64-mingw32/x64 (64-bit)
locale:
[1] LC_COLLATE=German_Germany.1252 LC_CTYPE=German_Germany.1252 LC_MONETARY=German_Germany.1252 LC_NUMERIC=C LC_TIME=German_Germany.1252
attached base packages:
[1] stats graphics grDevices datasets utils methods base
other attached packages:
[1] data.table_1.8.8 fortunes_1.5-0
Good question. By default in v1.8.8 :
> options()$datatable.alloccol
max(100, 2 * ncol(DT))
That's probably not the best default. Try changing it :
options(datatable.alloccol = quote(max(100L, ncol(DT)+64L))
UPDATE: I've now changed the default in v1.8.9 to that.
That option just controls how many spare column pointer slots are allocated so that :=
can add columns by reference.
From NOTES in NEWS for v1.8.9
- The default for
datatable.alloccol
has changed frommax(100L, 2L*ncol(DT))
tomax(100L, ncol(DT)+64L)
. And a pointer to?truelength
has been added to an error message as sugggested and thanks to Roland :
Potential problems from over-allocating truelength more than 1000 times
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With