[Rd] New R version - Issue with as.vector coercion on data.frame

[Martin Maechler]

>>>>> Bill Dunlap 
>>>>>     on Mon, 2 May 2022 08:07:41 -0700 writes:

    > In R-4.1.2 (and before) as.vector(aDataFrame) returned aDataFrame,
    > unchanged.  E.g.,
    4.1.2> aDataFrame <- data.frame(X=101:103, Y=201:203, Z=301:303)
    4.1.2> attr(aDataFrame, "anAttr") <- "an attribute"
    4.1.2> identical(as.vector(aDataFrame), aDataFrame)
    > [1] TRUE
    4.1.2> dput(aDataFrame)
    > structure(list(X = 101:103, Y = 201:203, Z = 301:303), class =
    > "data.frame", row.names = c(NA, -3L), anAttr = "an attribute")

    > In R-4.2.0 it

is based on the already existing  as.list.data.frame() which

    > removes the class, names, and row.names attributes, but
    > leaves other stray attributes alone.  E.g.,
    4.2.0> aDataFrame <- data.frame(X=101:103, Y=201:203, Z=301:303)
    4.2.0> attr(aDataFrame, "anAttr") <- "an attribute"
    4.2.0> identical(as.vector(aDataFrame), aDataFrame)
    > [1] FALSE
    4.2.0> dput(aDataFrame)
    > structure(list(X = 101:103, Y = 201:203, Z = 301:303), class =
    > "data.frame", row.names = c(NA, -3L), anAttr = "an attribute")
    4.2.0> dput(as.vector(aDataFrame))
    > structure(list(X = 101:103, Y = 201:203, Z = 301:303), anAttr = "an
    > attribute")

Yes, this was a somewhat "hard" change in an effort to slightly
improve the still unsatisfactory behavior of  as.vector() and is.vector().
and it solved a very long standing FIXME / TODO which wanted
that

   as.vector(df, mode = "list")    should be the same as
   as.list  (df)


   [....]

    > I think that as.vector() is a function that few people should use.   It has
    > almost nothing to do with with the notion of a vector in math or physics.

I tend to agree on that.

The confusion comes from the fact that in R (and S before it, no?)
a list() or an expression() are also a 'vector'  {a generalized one},

but that math/physics/... term of a "vector" is what is an
"atomic vector" in R.

Note that we also ventured into a relatively time confusing effort
to do more about the  as.vector()  <-->  is.vector()  decrepancies.

One step towards that I'd like to take up again --- and was in
the mean time helped by some package maintainers who adapted
their code --- is to make  is.atomic(.)  truly a test for the
above "atomic vector"s.
It currently is *not*, as   is.atomic(NULL)   is TRUE,
but NULL is really not an atomic vector.

So our plan has been that is.atomic(NULL) should become FALSE, so
programmeRs could use is.atomic() checks before using
as.vector(),  which would bring them back to the traditional
realm of 1st semester math where your vector spaces are
typically R^n (or C^n ..)

Martin


    > -Bill

    > On Mon, May 2, 2022 at 2:19 AM Marc Weibel <marc.weibel using eniso-partners.com>
    > wrote:

    >> Hi
    >> 
    >> I recently upgraded to the latest r version and got an issue with the
    >> command as.vector() on a data.frame object. With the previous R version the
    >> returned object was a vector but now it is formatted as a list (see below)
    >> 
    >> 
    >> > futures_quotes
    >> SMI Future SMIM Future Euro Stoxx 50 Future S&P 500 Mini Future FTSE 100
    >> Future Nikkei 225 Future
    >> 1      11999        2871                 3692                4150
    >> 7419             26853
    >> > class(futures_quotes)
    >> [1] "data.frame"
    >> 
    >> > as.vector(futures_quotes[1,1:3])
    >> $`SMI Future`
    >> [1] 11999
    >> 
    >> $`SMIM Future`
    >> [1] 2871
    >> 
    >> $`Euro Stoxx 50 Future`
    >> [1] 3692
    >> 
    >> > class(as.vector(futures_quotes[1,1:3]))
    >> [1] "list"
    >> 
    >> 
    >> Can you help me on this ? (I solved the issue temporarily with
    >> as.vector(t(futures_quotes[1,1:3]))
    >> 
    >> Kind regards
    >> Marc Weibel
    >> 
    >> 
    >> <http://www.enisopartners.com/>
    >> Dr. Marc Weibel
    >> CIO & Partner
    >> Tel: +41 44 286 17 02
    >> Mob: +41 78 892 96 55
    >> E-Mail: marc.weibel using enisopartners.com <mailto:
    >> marc.weibel using enisopartners.com>
    >> 
    >> Eniso Partners AG
    >> Claridenstrasse 34 | Postfach | 8022 Zürich
    >> 
    >> <https://www.linkedin.com/in/marc-weibel-29a396/>