seq improperly increments dates (PR#9120)

That's, in fact, the way seq.dates works in the chron package:

library(chron)
x <- chron("01/31/2006")
seq(x, by = "month", length = 2) # 01/31/06 02/28/06

See the help desk article in R News 4/1 for more about the main
date classes.

8/3/06, Ponzio, Stephen [CIB-LAVA] <stephen.ponzio (AT) citigroup (DOT) comwrote:
You're right, it's tricky.

I guess I would expect Jan. 30 + 1 month = Feb. 28.

couse, then Jan. 30 + 1 month = Jan. 28 + 1 month;
I understand.

Being that "1 month" is imprecise in terms of number of days,
this anomaly is preferable to skipping months (Jan. 30 + 1 month = March 2),
in my opinion.

That is what I expected, anyway.

Thanks,
Stephen
>
>
>
Message
From: Duncan Murdoch [mailto:murdoch (AT) stats (DOT) uwo.ca]
Sent: Thursday, August 03, 2006 11:24 AM
To: Ponzio, Stephen [CIB-LAVA]
Cc: r-devel (AT) stat (DOT) math.ethz.ch; R-bugs (AT) biostat (DOT) ku.dk
Subject: Re: [Rd] seq improperly increments dates (PR#9120)
--
8/3/2006 10:26 AM, stephen.ponzio (AT) citigroup (DOT) com wrote:
Full_Name: Stephen Ponzio
Version: 2.3.1
S: Windows
Submission from: (NULL) (199.67.138.42)
--
With the option by="1 month" and a date that is the 31st,
the function seq doesn't give the last day of successive months,
as I would expect it should:

Why would you expect that? I'd expect it to give the day 30 days after
the 1st, and that appears to be what it does. This doesn't look like a
bug to me.

I don't understand what logic you would like it to use. If I ask to
increment Jan 30 in a non-leap-year by one month, should I get March 2
(29 days after the 1st of January, as now), February 27 (the day before
the end of the month), or February 28 (the closest you can get in February)?

Whatever you choose, you'll get anomalies. Currently it skips months in
sequences like you show below. If you try for the same distance from
the end of the month each time, or truncate to the end of a short month,
you lose important arithmetic properties: x - y is not equal to
(x+1month) - (y+1month).

Duncan Murdoch
>
>
>Sys.Date()
[1] "2006-08-03"
>
>Sys.Date()-3
[1] "2006-07-31"

# WRKS K on Aug. 1:
>seq(Sys.Date()-2, len=5, by="1 month")
[1] "2006-08-01" "2006-09-01" "2006-10-01" "2006-11-01" "2006-12-01"

# DESN'T WRK on July 31 (list has no date in Sept):
>seq(Sys.Date()-3, len=5, by="1 month")
[1] "2006-07-31" "2006-08-31" "2006-10-01" "2006-10-31" "2006-12-01"

R-devel (AT) r-project (DOT) org mailing list

R-devel (AT) r-project (DOT) org mailing list

R-devel (AT) r-project (DOT) org mailing list

"Gabor" == Gabor Grothendieck <ggrothendieck (AT) gmail (DOT) com>
on Thu, 3 Aug 2006 20:14:24 -0400 writes:

GaborThat's, in fact, the way seq.dates works in the chron package:
Gaborlibrary(chron)
Gaborx <- chron("01/31/2006")
Gaborseq(x, by = "month", length = 2) # 01/31/06 02/28/06

Hmm, so, by "logic",
2006-01-31 + 1month |-2006-02-28
2006-01-30 + 1month |-2006-02-27 (?)
2006-01-29 + 1month |-2006-02-26 (?)
2006-01-28 + 1month |-2006-02-25 ??

I really don't like 'chron's behavior which seems much less
logical to me than what R does with the official "Date" objects :

options(width=88)
for(d in 28:31) print(seq(as.Date(paste("2006-01", d, sep="-")), len = 6, by="1 month"))
[1] "2006-01-28" "2006-02-28" "2006-03-28" "2006-04-28" "2006-05-28" "2006-06-28"
[1] "2006-01-29" "2006-03-01" "2006-03-29" "2006-04-29" "2006-05-29" "2006-06-29"
[1] "2006-01-30" "2006-03-02" "2006-03-30" "2006-04-30" "2006-05-30" "2006-06-30"
[1] "2006-01-31" "2006-03-03" "2006-03-31" "2006-05-01" "2006-05-31" "2006-07-01"

{which, BTW, *is* reproducible code; the bug report was only
reproducible on the day it was posted because it sillily used
Sys.date()}

GaborSee the help desk article in R News 4/1 for more about the main
Gabordate classes.

Gabor8/3/06, Ponzio, Stephen [CIB-LAVA] <stephen.ponzio (AT) citigroup (DOT) comwrote:
>You're right, it's tricky.
>
>I guess I would expect Jan. 30 + 1 month = Feb. 28.
>
>couse, then Jan. 30 + 1 month = Jan. 28 + 1 month;
>I understand.
>
>Being that "1 month" is imprecise in terms of number of days,
>this anomaly is preferable to skipping months (Jan. 30 + 1 month = March 2),
>in my opinion.
>
>That is what I expected, anyway.
>
>Thanks,
>Stephen

[]

R-devel (AT) r-project (DOT) org mailing list

Here are some more seq.dates examples from chron.
It seems that if seq.Date cannot output day of the month n
(because the month has fewer days) then it outputs n days from
the start of the month even if that goes into the next month
whereas seq.dates in chron outputs the end of the month in that case.

library(chron)
seq(chron("01/31/2006"), by = "month", length = 6)
[1] 01/31/06 02/28/06 03/31/06 04/30/06 05/31/06 06/30/06

seq(chron("01/30/2006"), by = "month", length = 6)
[1] 01/30/06 02/28/06 03/30/06 04/30/06 05/30/06 06/30/06

seq(chron("01/29/2006"), by = "month", length = 6)
[1] 01/29/06 02/28/06 03/29/06 04/29/06 05/29/06 06/29/06

seq(chron("01/28/2006"), by = "month", length = 6)
[1] 01/28/06 02/28/06 03/28/06 04/28/06 05/28/06 06/28/06

seq(chron("01/27/2006"), by = "month", length = 6)
[1] 01/27/06 02/27/06 03/27/06 04/27/06 05/27/06 06/27/06

8/4/06, Martin Maechler <maechler (AT) stat (DOT) math.ethz.chwrote:
"Gabor" == Gabor Grothendieck <ggrothendieck (AT) gmail (DOT) com>
on Thu, 3 Aug 2006 20:14:24 -0400 writes:

GaborThat's, in fact, the way seq.dates works in the chron package:
Gaborlibrary(chron)
Gaborx <- chron("01/31/2006")
Gaborseq(x, by = "month", length = 2) # 01/31/06 02/28/06

Hmm, so, by "logic",
2006-01-31 + 1month |-2006-02-28
2006-01-30 + 1month |-2006-02-27 (?)
2006-01-29 + 1month |-2006-02-26 (?)
2006-01-28 + 1month |-2006-02-25 ??

I really don't like 'chron's behavior which seems much less
logical to me than what R does with the official "Date" objects :

options(width=88)
for(d in 28:31) print(seq(as.Date(paste("2006-01", d, sep="-")), len = 6, by="1 month"))
[1] "2006-01-28" "2006-02-28" "2006-03-28" "2006-04-28" "2006-05-28" "2006-06-28"
[1] "2006-01-29" "2006-03-01" "2006-03-29" "2006-04-29" "2006-05-29" "2006-06-29"
[1] "2006-01-30" "2006-03-02" "2006-03-30" "2006-04-30" "2006-05-30" "2006-06-30"
[1] "2006-01-31" "2006-03-03" "2006-03-31" "2006-05-01" "2006-05-31" "2006-07-01"
--
{which, BTW, *is* reproducible code; the bug report was only
reproducible on the day it was posted because it sillily used
Sys.date()}

GaborSee the help desk article in R News 4/1 for more about the main
Gabordate classes.

Gabor8/3/06, Ponzio, Stephen [CIB-LAVA] <stephen.ponzio (AT) citigroup (DOT) comwrote:
>You're right, it's tricky.
>>
>I guess I would expect Jan. 30 + 1 month = Feb. 28.
>>
>couse, then Jan. 30 + 1 month = Jan. 28 + 1 month;
>I understand.
>>
>Being that "1 month" is imprecise in terms of number of days,
>this anomaly is preferable to skipping months (Jan. 30 + 1 month = March 2),
>in my opinion.
>>
>That is what I expected, anyway.
>>
>Thanks,
>Stephen
>
[]

R-devel (AT) r-project (DOT) org mailing list

8/4/2006 4:18 AM, Martin Maechler wrote:
"Gabor" == Gabor Grothendieck <ggrothendieck (AT) gmail (DOT) com>
on Thu, 3 Aug 2006 20:14:24 -0400 writes:

GaborThat's, in fact, the way seq.dates works in the chron package:
Gaborlibrary(chron)
Gaborx <- chron("01/31/2006")
Gaborseq(x, by = "month", length = 2) # 01/31/06 02/28/06

Hmm, so, by "logic",
2006-01-31 + 1month |-2006-02-28
2006-01-30 + 1month |-2006-02-27 (?)
2006-01-29 + 1month |-2006-02-26 (?)
2006-01-28 + 1month |-2006-02-25 ??

I really don't like 'chron's behavior which seems much less
logical to me than what R does with the official "Date" objects :

>options(width=88)
>for(d in 28:31) print(seq(as.Date(paste("2006-01", d, sep="-")), len = 6, by="1 month"))
[1] "2006-01-28" "2006-02-28" "2006-03-28" "2006-04-28" "2006-05-28" "2006-06-28"
[1] "2006-01-29" "2006-03-01" "2006-03-29" "2006-04-29" "2006-05-29" "2006-06-29"
[1] "2006-01-30" "2006-03-02" "2006-03-30" "2006-04-30" "2006-05-30" "2006-06-30"
[1] "2006-01-31" "2006-03-03" "2006-03-31" "2006-05-01" "2006-05-31" "2006-07-01"

{which, BTW, *is* reproducible code; the bug report was only
reproducible on the day it was posted because it sillily used
Sys.date()}

I think the obvious solution here is to ask R Core to move the earth
slightly closer to the sun, so the year is exactly 360 days long, and
these problems don't arise.

perhaps exactly 372 days would be better; that would be further from
the sun and would also help with global warming.

Duncan Murdoch

GaborSee the help desk article in R News 4/1 for more about the main
Gabordate classes.

Gabor8/3/06, Ponzio, Stephen [CIB-LAVA] <stephen.ponzio (AT) citigroup (DOT) comwrote:
>You're right, it's tricky.
>
>I guess I would expect Jan. 30 + 1 month = Feb. 28.
>
>couse, then Jan. 30 + 1 month = Jan. 28 + 1 month;
>I understand.
>
>Being that "1 month" is imprecise in terms of number of days,
>this anomaly is preferable to skipping months (Jan. 30 + 1 month = March 2),
>in my opinion.
>
>That is what I expected, anyway.
>
>Thanks,
>Stephen

[]

R-devel (AT) r-project (DOT) org mailing list

R-devel (AT) r-project (DOT) org mailing list

Duncan Murdoch wrote:
8/4/2006 4:18 AM, Martin Maechler wrote:
"Gabor" == Gabor Grothendieck <ggrothendieck (AT) gmail (DOT) com>
on Thu, 3 Aug 2006 20:14:24 -0400 writes:
>GaborThat's, in fact, the way seq.dates works in the chron package:
>Gaborlibrary(chron)
>Gaborx <- chron("01/31/2006")
>Gaborseq(x, by = "month", length = 2) # 01/31/06 02/28/06
>>
>Hmm, so, by "logic",
>2006-01-31 + 1month |-2006-02-28
>2006-01-30 + 1month |-2006-02-27 (?)
>2006-01-29 + 1month |-2006-02-26 (?)
>2006-01-28 + 1month |-2006-02-25 ??
>>
>I really don't like 'chron's behavior which seems much less
>logical to me than what R does with the official "Date" objects :
>>
options(width=88)
for(d in 28:31) print(seq(as.Date(paste("2006-01", d, sep="-")), len = 6, by="1 month"))
>[1] "2006-01-28" "2006-02-28" "2006-03-28" "2006-04-28" "2006-05-28" "2006-06-28"
>[1] "2006-01-29" "2006-03-01" "2006-03-29" "2006-04-29" "2006-05-29" "2006-06-29"
>[1] "2006-01-30" "2006-03-02" "2006-03-30" "2006-04-30" "2006-05-30" "2006-06-30"
>[1] "2006-01-31" "2006-03-03" "2006-03-31" "2006-05-01" "2006-05-31" "2006-07-01"
>>
>{which, BTW, *is* reproducible code; the bug report was only
>reproducible on the day it was posted because it sillily used
>Sys.date()}

I think the obvious solution here is to ask R Core to move the earth
slightly closer to the sun, so the year is exactly 360 days long, and
these problems don't arise.

perhaps exactly 372 days would be better; that would be further from
the sun and would also help with global warming.

Duncan Murdoch

I am afraid that moving the earty so far out to induce an ice age, so
I would please ask R-core to leave the earth alone!

Kjetil


>GaborSee the help desk article in R News 4/1 for more about the main
>Gabordate classes.
>>
>Gabor8/3/06, Ponzio, Stephen [CIB-LAVA] <stephen.ponzio (AT) citigroup (DOT) comwrote:
>>You're right, it's tricky.
>>
>>I guess I would expect Jan. 30 + 1 month = Feb. 28.
>>
>>couse, then Jan. 30 + 1 month = Jan. 28 + 1 month;
>>I understand.
>>
>>Being that "1 month" is imprecise in terms of number of days,
>>this anomaly is preferable to skipping months (Jan. 30 + 1 month = March 2),
>>in my opinion.
>>
>>That is what I expected, anyway.
>>
>>Thanks,
>>Stephen
>>
>[]
>>
>
>R-devel (AT) r-project (DOT) org mailing list
>

R-devel (AT) r-project (DOT) org mailing list

R-devel (AT) r-project (DOT) org mailing list