Re: Split apply combine in data.table

В Wed, 4 Feb 2026 11:56:37 +0000
Naresh Gurbuxani <[email protected]> пишет:

> myapprox <- mydt[, list(yfunc = list(approxfun(x, y, rule = 2))), by =
> list(date)]

The grouping operations in data.table reuse the same objects 'x', 'y'
with different contents when evaluating the expression for different
groups. As a result, the same objects end up being captured by
approxfun() for different groups:

myapprox$yfunc |> sapply(\(.) environment(.)$x |> address()) |> unique()
# [1] "0x562038fdbd88" # <-- only one address instead of three

Ideally, data.table should notice that its internal object is captured
instead of blindly reusing it for the next group, but it doesn't
currently do that. As a workaround, you can copy the otherwise-shared
objects explicitly:

myapprox <- mydt[, 
 .(yfunc = .(approxfun(copy(x), copy(y), rule = 2))),
 by = .(date)
]
myapprox$yfunc |> sapply(\(.) environment(.)$x |> address()) |> unique()
# [1] "0x562039b93238" "0x562039b92d38" "0x562039b92888"

-- 
Best regards,
Ivan