Re: [Rd] head.ts, tail.ts loses time
> Gabor Grothendieck
> on Tue, 11 Jun 2024 09:13:49 -0400 writes:
> It isn't really clear that it can't work. This does work by inserting
NA's.
> library(zoo)
> as.ts(as.zoo(lynx)[ c(1:3, 7) ] )
> ## Time Series:
> ## Start = 1821
> ## End = 1827
> ## Frequency = 1
> ## [1] 269 321 585 NA NA NA 3928
You are right, Gabor, such an implementation of `[.ts`
*would* make sense, too.
But given that head.ts() and tail.ts() -- slightly compactified
from Spencer's proposal, are now simple and "robust",
I did not want to make such a "strong" change to such a basic class and its `[`
operator.
(I'm not *against* it either currently, but I'm not convinced
it's worth the effort with possible subsequent changes needed
in code which has relied on the old behavior for > 30 years
if you count the pre-R S version, too.)
Best, Martin
> On Mon, Jun 10, 2024 at 10:32 AM Martin Maechler
> wrote:
>>
>> > Spencer Graves
>> > on Mon, 10 Jun 2024 07:50:13 -0500 writes:
>>
>> > Hi, Gabor et al.: Thanks for this. I should change my
>> > current application to use either zoo or xts, as Gabor
>> > suggests.
>>
>>
>> > However, I was surprised to learn that "[.ts" does NOT
>> > return an object of class "ts". I see that "head.default"
>> > and "head.matrix" both call "[", so "head" cannot return a
>> > ts object, because "[" doesn't.
>>
>> Yes, the default head() and tail() are built on `[` very much
>> on purpose.
>> Note that `[` should *not* keep the "ts" property in
>> general, e.g.,
>> lynx[c(1:3, 7)]
>> cannot be a regular time series
>>
>> I think I'd consider using windows() for a head.ts() and tail.ts(),
>> but in any case, I am sympathetic adding such methods to "base R"'s
>> utils package.
>>
>>
>> Martin
>>
>> > Best Wishes, Spencer Graves
>>
>>
>> > On 6/9/24 8:40 PM, Gabor Grothendieck wrote:
>> >> zoo overcomes many of the limitations of ts:
>> >>
>> >> library(zoo) as.ts(head(as.zoo(presidents))) ## Qtr1 Qtr2
>> >> Qtr3 Qtr4 ## 1945 NA 87 82 75 ## 1946 63 50
>> >>
>> >> xts also works here.
>> >>
>> >> On Sun, Jun 9, 2024 at 12:04 PM Spencer Graves
>> >> wrote:
>> >>>
>> >>> Hello, All:
>> >>>
>> >>>
>> >>> The 'head' and 'tail' functions strip the time from a
>> >>> 'ts' object. Example:
>> >>>
>> >>>
>> >>> > head(presidents) [1] NA 87 82 75 63 50
>> >>>
>> >>>
>> >>> > window(presidents, 1945, 1946.25) Qtr1 Qtr2 Qtr3 Qtr4
>> >>> 1945 NA 87 82 75 1946 63 50
>> >>>
>> >>>
>> >>> Below please find code for 'head.ts' and 'tail.ts' that
>> >>> matches 'window'.
>> >>>
>> >>>
>> >>> Comments? Spencer Graves
>> >>>
>> >>> head.ts <- function(x, n=6L, ...){ tmx <-
>> >>> as.numeric(time(x))
>> >>> #
>> >>> utils:::checkHT(n, d <- dim(x)) if(is.na(n[1]) ||
>> >>> n[1]==0)ts(NULL)
>> >>> #
>> >>> firstn <- head(tmx, n[1]) if(is.null(d)){
>> >>> return(window(x, firstn[1], tail(firstn, 1))) } else{
>> >>> if(length(n)<2){ return(window(x, firstn[1],
>> >>> tail(firstn, 1))) } else { Cols <- head(1:d[2], n[2])
>> >>> xn2 <- x[, Cols[1]:tail(Cols, 1)] return(window(xn2,
>> >>> firstn[1], tail(firstn, 1))) } } }
>> >>>
>> >>>
>> >>> tail.ts <- function (x, n = 6L, ...) {
>> >>> utils:::checkHT(n, d <- dim(x)) tmx <-
>> >>> as.numeric(time(x))
>> >>> #
>> >>> if(is.na(n[1]) || n[1]==0)ts(NULL)
>> >>> #
>> >>> lastn <- tail(tmx, n[1]) if(is.null(d)){
>> >>> return(window(x, lastn[1], tail(lastn, 1))) } else{
>> >>> if(length(n)<2){ return(window(x, lastn[1], tail(lastn,
>> >>> 1))) } else { Cols <- head(1:d[2], n[2]) xn2 <- x[,
>> >>> Cols[1]:tail(Cols, 1)] return(window(xn2, lastn[1],
>> >>> tail(lastn, 1))) } } }
>> >>>
>> >>>
>> >>> # examples head(presidents)
>> >>>
>> >>> head(presidents, 2)
>> >>>
>> >>> npresObs <- length(presidents) head(presidents,
>> >>> 6-npresObs)
>> >>>
>> >>> try(head(presidents, 1:2)) # 'try-error'
>> >>>
>> >>> try(head(presidents, 0)) # 'try-error'
>> >>>
>> >>> # matrix time series str(pres <-
>> >>> cbind(n=1:length(presidents), presidents)) head(pres, 2)
>> >>>
>> >>> head(pres, 2-npresObs)
>> >>>
>> >>> head(pres, 1:2) head(pres, 2:1) head(pres, 1:3)
>> >>>
>> >>> # examples tail(presidents)
>> >>>
>> >>> tail(presidents, 2)
>> >>>
>> >>> npresObs <- length(presidents) tail(presidents,
>> >>> 6-npresObs)
>> >>>
>> >>> try(tail(presidents, 1:2)) # 'try-error'
>> >>>
>> >>> try(tail(presidents, 0)) # 'try-error'
>> >>>
>> >>> # matrix time series str(pres <-
>> >>> cbind(n=1:length(presidents), presidents
Re: [Rd] head.ts, tail.ts loses time
> Spencer Graves
> on Mon, 10 Jun 2024 09:45:46 -0500 writes:
> Hi, Martin et al.:
> On 6/10/24 9:32 AM, Martin Maechler wrote:
>>> Spencer Graves
>>> on Mon, 10 Jun 2024 07:50:13 -0500 writes:
>>
>> > Hi, Gabor et al.: Thanks for this. I should change my
>> > current application to use either zoo or xts, as Gabor
>> > suggests.
>>
>>
>> > However, I was surprised to learn that "[.ts" does NOT
>> > return an object of class "ts". I see that "head.default"
>> > and "head.matrix" both call "[", so "head" cannot return a
>> > ts object, because "[" doesn't.
>>
>> Yes, the default head() and tail() are built on `[` very much
>> on purpose.
>> Note that `[` should *not* keep the "ts" property in
>> general, e.g.,
>> lynx[c(1:3, 7)]
>> cannot be a regular time series
> Agreed.
>>
>> I think I'd consider using windows() for a head.ts() and tail.ts(),
>> but in any case, I am sympathetic adding such methods to "base R"'s
>> utils package.
> The code I provided below for head.ts() and tail.ts() does that: I
> took the code for head.default and head.matrix, etc., computed tmx <-
> as.numeric(time(x)), and then used head(tmx) [and tail(tmx)] in
"window()".
Indeed. I've found that the new methods really belong to pkg
'stats' (where "ts" are), and hence renamed and exported the internal
.checkHT(), and shence the change became somewhat more extensive:
r86728 | maechler | 2024-06-13 10:36:51 +0200 (Thu, 13 Jun 2024) | 1 line
Changed paths:
M doc/NEWS.Rd
M src/library/stats/NAMESPACE
M src/library/stats/R/ts.R
M src/library/stats/man/ts.Rd
M src/library/utils/NAMESPACE
M src/library/utils/R/head.R
M src/library/utils/man/head.Rd
M tests/Examples/stats-Ex.Rout.save
add head() & tail() methods for "ts"(time series) ==> export .checkHT() utility
With thanks to Spencer Graves,
Martin
> Thanks for your reply.
> sg
>>
>>
>> Martin
>>
>> > Best Wishes, Spencer Graves
>>
>>
>> > On 6/9/24 8:40 PM, Gabor Grothendieck wrote:
>> >> zoo overcomes many of the limitations of ts:
>> >>
>> >> library(zoo) as.ts(head(as.zoo(presidents))) ## Qtr1 Qtr2
>> >> Qtr3 Qtr4 ## 1945 NA 87 82 75 ## 1946 63 50
>> >>
>> >> xts also works here.
>> >>
>> >> On Sun, Jun 9, 2024 at 12:04 PM Spencer Graves
>> >> wrote:
>> >>>
>> >>> Hello, All:
>> >>>
>> >>>
>> >>> The 'head' and 'tail' functions strip the time from a
>> >>> 'ts' object. Example:
>> >>>
>> >>>
>> >>> > head(presidents) [1] NA 87 82 75 63 50
>> >>>
>> >>>
>> >>> > window(presidents, 1945, 1946.25) Qtr1 Qtr2 Qtr3 Qtr4
>> >>> 1945 NA 87 82 75 1946 63 50
>> >>>
>> >>>
>> >>> Below please find code for 'head.ts' and 'tail.ts' that
>> >>> matches 'window'.
>> >>>
>> >>>
>> >>> Comments? Spencer Graves
>> >>>
>> >>> head.ts <- function(x, n=6L, ...){ tmx <-
>> >>> as.numeric(time(x))
>> >>> #
>> >>> utils:::checkHT(n, d <- dim(x)) if(is.na(n[1]) ||
>> >>> n[1]==0)ts(NULL)
>> >>> #
>> >>> firstn <- head(tmx, n[1]) if(is.null(d)){
>> >>> return(window(x, firstn[1], tail(firstn, 1))) } else{
>> >>> if(length(n)<2){ return(window(x, firstn[1],
>> >>> tail(firstn, 1))) } else { Cols <- head(1:d[2], n[2])
>> >>> xn2 <- x[, Cols[1]:tail(Cols, 1)] return(window(xn2,
>> >>> firstn[1], tail(firstn, 1))) } } }
>> >>>
>> >>>
>> >>> tail.ts <- function (x, n = 6L, ...) {
>> >>> utils:::checkHT(n, d <- dim(x)) tmx <-
>> >>> as.numeric(time(x))
>> >>> #
>> >>> if(is.na(n[1]) || n[1]==0)ts(NULL)
>> >>> #
>> >>> lastn <- tail(tmx, n[1]) if(is.null(d)){
>> >>> return(window(x, lastn[1], tail(lastn, 1))) } else{
>> >>> if(length(n)<2){ return(window(x, lastn[1], tail(lastn,
>> >>> 1))) } else { Cols <- head(1:d[2], n[2]) xn2 <- x[,
>> >>> Cols[1]:tail(Cols, 1)] return(window(xn2, lastn[1],
>> >>> tail(lastn, 1))) } } }
>> >>>
>> >>>
>> >>> # examples head(presidents)
>> >>>
>> >>> head(presidents, 2)
>> >>>
>> >>> npresObs <- length(presidents) head(presidents,
>> >>> 6-npresObs)
>> >>>
>> >>> try(head(presidents, 1:2)) # 'try-error'
>> >>>
>> >>> try(head(presidents, 0)) # 'try-error'
>> >>>
>> >>> # matrix time series str(pres <-
>> >>> cbind(n=1:length(presidents), presidents)) head(pres, 2)
>> >>>
>> >>> head(pres, 2-npresObs)
>> >>>
>> >>> head(pres, 1:2) head(pres, 2:1) head(pres, 1:3)
>> >>>
>> >>> # examples tail(presidents)
>> >>>
>> >>> tail(presidents, 2)
>> >>>
>> >>> npresObs <- length(
Re: [Rd] head.ts, tail.ts loses time
> It isn't really clear that it can't work. This does work by inserting NA's...
>
> library(zoo)
> as.ts(as.zoo(lynx)[ c(1:3, 7) ] )
If by 'this' you mean indexing, it would be very confusing and error prone for
expressions like lynx[c(1:3, 7)] (lynx is from class 'ts') to return a ts
object with NA's inserted and, even more so, since this has been unambiguously
documented for ages in '?ts'. For 'zoo' objects, the situation is different
since they have a time index by definition, so the above index doesn't
introduce artificial NA's:
> as.zoo(lynx)[ c(1:3, 7) ]
1821 1822 1823 1827
269 321 585 3928
> coredata(as.zoo(lynx)[ c(1:3, 7) ])
[1] 269 321 585 3928
On the other hand, 'ts' methods for 'head' and 'tail' would be suitable, since
the indexing is contiguous by definition there. Also, compatibility problems
may not be a big concern for these functions.
Georgi Boshnakov
Date: Tue, 11 Jun 2024 09:13:49 -0400
From: Gabor Grothendieck
To: Martin Maechler
Cc: Spencer Graves , r-devel
Subject: Re: [Rd] head.ts, tail.ts loses time
Message-ID:
Content-Type: text/plain; charset="utf-8"
It isn't really clear that it can't work. This does work by inserting NA's...
library(zoo)
as.ts(as.zoo(lynx)[ c(1:3, 7) ] )
## Time Series:
## Start = 1821
## End = 1827
## Frequency = 1
## [1] 269 321 585 NA NA NA 3928
On Mon, Jun 10, 2024 at 10:32 AM Martin Maechler
wrote:
>
> > Spencer Graves
> > on Mon, 10 Jun 2024 07:50:13 -0500 writes:
>
> > Hi, Gabor et al.: Thanks for this. I should change my
> > current application to use either zoo or xts, as Gabor
> > suggests.
>
>
> > However, I was surprised to learn that "[.ts" does NOT
> > return an object of class "ts". I see that "head.default"
> > and "head.matrix" both call "[", so "head" cannot return a
> > ts object, because "[" doesn't.
>
> Yes, the default head() and tail() are built on `[` very much
> on purpose.
> Note that `[` should *not* keep the "ts" property in
> general, e.g.,
> lynx[c(1:3, 7)]
> cannot be a regular time series
>
> I think I'd consider using windows() for a head.ts() and tail.ts(),
> but in any case, I am sympathetic adding such methods to "base R"'s
> utils package.
>
>
> Martin
>
> > Best Wishes, Spencer Graves
>
>
> > On 6/9/24 8:40 PM, Gabor Grothendieck wrote:
> >> zoo overcomes many of the limitations of ts:
> >>
> >> library(zoo) as.ts(head(as.zoo(presidents))) ## Qtr1 Qtr2
> >> Qtr3 Qtr4 ## 1945 NA 87 82 75 ## 1946 63 50
> >>
> >> xts also works here.
> >>
> >> On Sun, Jun 9, 2024 at 12:04 PM Spencer Graves
> >> wrote:
> >>>
> >>> Hello, All:
> >>>
> >>>
> >>> The 'head' and 'tail' functions strip the time from a
> >>> 'ts' object. Example:
> >>>
> >>>
> >>> > head(presidents) [1] NA 87 82 75 63 50
> >>>
> >>>
> >>> > window(presidents, 1945, 1946.25) Qtr1 Qtr2 Qtr3 Qtr4
> >>> 1945 NA 87 82 75 1946 63 50
> >>>
> >>>
> >>> Below please find code for 'head.ts' and 'tail.ts' that
> >>> matches 'window'.
> >>>
> >>>
> >>> Comments? Spencer Graves
> >>>
> >>> head.ts <- function(x, n=6L, ...){ tmx <-
> >>> as.numeric(time(x))
> >>> #
> >>> utils:::checkHT(n, d <- dim(x)) if(is.na(n[1]) ||
> >>> n[1]==0)ts(NULL)
> >>> #
> >>> firstn <- head(tmx, n[1]) if(is.null(d)){
> >>> return(window(x, firstn[1], tail(firstn, 1))) } else{
> >>> if(length(n)<2){ return(window(x, firstn[1],
> >>> tail(firstn, 1))) } else { Cols <- head(1:d[2], n[2])
> >>> xn2 <- x[, Cols[1]:tail(Cols, 1)] return(window(xn2,
> >>> firstn[1], tail(firstn, 1))) } } }
> >>>
> >>>
> >>> tail.ts <- function (x, n = 6L, ...) {
> >>> utils:::checkHT(n, d <- dim(x)) tmx <-
> >>> as.numeric(time(x))
> >>> #
> >>> if(is.na(n[1]) || n[1]==0)ts(NULL)
> >>> #
> >>> lastn <- tail(tmx, n[1]) if(is.null(d)){
> >>> return(window(x, lastn[1], tail(lastn, 1))) } else{
> >>> if(length(n)<2){ return(window(x, lastn[1], tail(lastn,
> >>> 1))) } else { Cols <- head(1:d[2], n[2]) xn2 <- x[,
> >>> Cols[1]:tail(Cols, 1)] return(window(xn2, lastn[1],
> >>> tail(lastn, 1))) } } }
> >>>
> >>>
> >>> # examples head(presidents)
> >>>
> >>> head(presidents, 2)
> >>>
> >>> npresObs <- length(presidents) head(presidents,
> >>> 6-npresObs)
> >>>
> >>> try(head(presidents, 1:2)) # 'try-error'
> >>>
> >>> try(head(presidents, 0)) # 'try-error'
> >>>
> >>> # matrix time series str(pres <-
> >>> cbind(n=1:length(presidents), presidents)) head(pres, 2)
> >>>
> >>> head(pres, 2-npresObs)
> >>>
> >>> head(pres, 1:2) head(pres, 2:1) head(pres, 1:3)
> >>>
> >>> # examples tail(presidents)
> >>>
> >>> tail(presidents, 2)
> >>>
> >>> npresObs <- length(president
[Rd] R-devel on Windows temporarily broken?
I had a very routine CI job fail twice this morning on r-devel on Windows; the package (in fine standard form) doesn't even install under win-builder r-devel. Whereas on Linux with r86731 everything is peachy. Dirk -- dirk.eddelbuettel.com | @eddelbuettel | e...@debian.org __ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
Re: [Rd] R-devel on Windows temporarily broken?
On 13 June 2024 at 15:26, Martin Maechler wrote: | > Dirk Eddelbuettel | > on Thu, 13 Jun 2024 07:20:00 -0500 writes: | | > I had a very routine CI job fail twice this morning on r-devel on Windows; | > the package (in fine standard form) doesn't even install under win-builder | > r-devel. Whereas on Linux with r86731 everything is peachy. | | > Dirk | | There was a texinfo typo for a while which Prof Brian Ripley | fixed in r86729. | | Hence r86731 is fine on Windows, too, at least when I look at | the R contributor svn dashboard (provided thanks to Jeroen Ooms): | https://contributor.r-project.org/svn-dashboard/ This ran twice, and blew up both times in a spot where it is 'impossible to blow up where it did' (line 183) https://github.com/TileDB-Inc/TileDB-R/actions/runs/9498253355/job/26179179289 Note that r-release ran fine on the same commit. Ditto with this win-builder r-devel build _not even installing_ https://win-builder.r-project.org/SNSy1J8W8H8b The package now also errors on r-devel-linux-x86_64-debian-gcc and was fine days ago (as we just uploaded that release last week). Neither one of these errors has, as best as I can tell, anything to do with texinfo, as you brought that up. Anyway, I just disabled r-devel in the CI matrix for now and will revisit in a few days. As I wrote, it works here (as I wrote) under '2024-06-13 r86731'. Dirk -- dirk.eddelbuettel.com | @eddelbuettel | e...@debian.org __ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
Re: [Rd] R-devel on Windows temporarily broken?
> Dirk Eddelbuettel > on Thu, 13 Jun 2024 07:20:00 -0500 writes: > I had a very routine CI job fail twice this morning on r-devel on Windows; > the package (in fine standard form) doesn't even install under win-builder > r-devel. Whereas on Linux with r86731 everything is peachy. > Dirk There was a texinfo typo for a while which Prof Brian Ripley fixed in r86729. Hence r86731 is fine on Windows, too, at least when I look at the R contributor svn dashboard (provided thanks to Jeroen Ooms): https://contributor.r-project.org/svn-dashboard/ __ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
