52

Why does iterating through a Date or POSIXct object result in numeric? For example:

test = as.Date("2009-01-01")
print( class( test ) )
# [1] "Date"
for ( day in test )
{
    print( class( day ) )
}
# [1] "numeric"

The same thing happens with POSIXct:

test = as.POSIXct("2009-01-01")
print( class( test ) )
# [1] "POSIXct" "POSIXt"
for ( day in test )
{
    print( class( day ) )
}
# [1] "numeric"
Henrik
  • 65,555
  • 14
  • 143
  • 159
Suraj
  • 35,905
  • 47
  • 139
  • 250

7 Answers7

39

?"for" says that seq (the part after in) is "[A]n expression evaluating to a vector (including a list and an expression) or to a pairlist or 'NULL'".

So your Date vector is being coerced to numeric because Date objects aren't strictly vectors:

is.vector(Sys.Date())
# [1] FALSE
is.vector(as.numeric(Sys.Date()))
# [1] TRUE

The same is true for POSIXct vectors:

is.vector(Sys.time())
# [1] FALSE
is.vector(as.numeric(Sys.time()))
# [1] TRUE
Joshua Ulrich
  • 173,410
  • 32
  • 338
  • 418
  • 2
    Pretty close, but I'm afraid that but there is no coercion in a for loop here, see [this answer](http://stackoverflow.com/a/23278464/3309529)..... – gagolews Apr 24 '14 at 19:52
  • 2
    @gagolews: That's a very subtle distinction that few people will understand or appreciate. Yes, `for` doesn't strip attributes and call `coerceVector` on the iterator (and potentially create a copy). It simply ignores the attributes. Regardless, the practical effect is the same. Had I said "treated as" instead of "coerced to", you wouldn't have had a point to make. – Joshua Ulrich Apr 24 '14 at 20:11
  • 1
    It would be less confusing to admit that Date objects _are_ "vectors" in the sense that most people understand the term, but with an attribute which is lost. For instance, `is.atomic(SysDate())` returns `TRUE`. The really surprising notion for me anyway was that for-loops will traverse lists. – IRTFM Apr 24 '14 at 20:15
  • 1
    @BondedDust, that is probably because if you do `is.vector(as.list(as.Date("2009-01-01")))` it will return `TRUE` – David Arenburg Apr 24 '14 at 20:17
  • 1
    So that's the reason why it's sometimes good to study the source code - the manual does not explain all the nuances. Indeed, as Joushua mentioned, from a plain user's point of view this is not so important. But it's good to have some insight. – gagolews Apr 24 '14 at 20:20
  • 3
    @gagolews: And don't misunderstand my comment, it's an interesting point, but I just don't think most people will care. ;) – Joshua Ulrich Apr 24 '14 at 20:21
24

loop through days (strings):

     days <- seq(from=as.Date('2011-02-01'), to=as.Date("2011-03-02"),by='days' )
     for ( i in seq_along(days) )
     {
          print(i)
           print(days[i])
      }
pleo
  • 241
  • 2
  • 2
19

You are not choosing the right function to apply to Date vectors when using for-loops. Better would be wrapping seq_along for pretty much every date or factor that is being looped across. Then you will do two thing: a) set it up so you are expecting an index that starts at 1, and b) protect against strange things that occur with zero length vectors. I also think it would be better to use it with factors, which the for-loops will turn into character vectors.

With reference to Joshua's answer (which is certainly correct and helpful), I think the is.vector function is a bit mislabeled or maybe just misunderstood. It could be more accurately be called hasNoAttributesOtherThanName. The property that most people consider "vectoric" is tested with is.atomic and Date and POSIXct objects will return TRUE from that test.

Henrik
  • 65,555
  • 14
  • 143
  • 159
IRTFM
  • 258,963
  • 21
  • 364
  • 487
  • 3
    +1 especially for `hasNoAttributesOtherThanName`, though I think it should be named `has_no_attributes_other_than_name`. ;-) – Joshua Ulrich Jan 05 '12 at 22:42
  • Your 'a)' point is valid, but regarding 'b)', the docs for `for` say "If `seq` has length zero the body of the loop is skipped." – Ken Williams Sep 06 '12 at 19:00
  • In this case it might not make a difference but in the case what the person was using 1:length(x) they get iterations that they should not want to get. Safer to use seq_along(). – IRTFM Sep 06 '12 at 19:21
9

It seems that the C function that implements the for loop does not copy any of the vector's attributes. This also includes the class attribute, which should make i appear to be a Date object.

You may study the source code of the do_for(SEXP, SEXP, SEXP, SEXP) function (the one called by R's for) here.

gagolews
  • 12,836
  • 2
  • 50
  • 75
6

It is an old question, but I am a novice on R and faced the same problem. Since my problem would be processed in parallel, I used foreach and saw the behaviour is different when compared with the normal for:

library(foreach)

start_date = as.Date("2013-08-1")
end_date = as.Date("2013-08-13")
days = seq(start_date, end_date, by = "day")

foreach(day = days, .combine='rbind') %dopar% {
  print(class(day))
}

[1] "Date"
[1] "Date"
[1] "Date"
[1] "Date"
...

As I am not experienced with the inner things of most of R, thus I do not know why foreach ends up having a different behaviour but that worked for my purpose, and hopefully might be useful for someone else.

Eduardo
  • 4,282
  • 2
  • 49
  • 63
1

Any numerical operation on date objects generally returns the number of days. In this, you are asking it to provide you with the number of days from the epoch. 14245 which is the number of days between 1970-01-01 - 2009-01-01

From ?Dates:

Dates are represented as the number of days since 1970-01-01, with negative values for earlier dates. They are always printed following the rules of the current Gregorian calendar, even though that calendar was not in use long ago (it was adopted in 1752 in Great Britain and its colonies).

It is intended that the date should be an integer, but this is not enforced in the internal representation. Fractional days will be ignored when printing. It is possible to produce fractional days via the mean method or by adding or subtracting (see Ops.Date).

Try adding print(day) to see what I mean. 

test = as.Date("2009-01-01")
print( class( test ) )
for ( day in test )
{
  print(day)
  print( class( day ) )
}
Brandon Bertelsen
  • 43,807
  • 34
  • 160
  • 255
  • 4
    hmm...which numeric operation am I performing? I'm simply using a for loop. If I use a for loop from i:length(test) and then call test[i] then I get a Date. Its not intuitive to me why the "for-each" would result in numeric – Suraj Jun 22 '11 at 04:05
  • 1
    If you want to loop through the number of days, use `for(1:as.numeric(test))` – Brandon Bertelsen Jun 22 '11 at 04:08
  • 1
    Your present for statement is just returning the number of days between test and 1970-01-01. Or, `day:test` but what you really want is something like `1:day:test` – Brandon Bertelsen Jun 22 '11 at 04:10
  • 5
    I guess I'm just not seeing it. Correct me if I'm wrong, but I read "for ( day in test )" as "iterate through each element in test and assign the value to variable day" – Suraj Jun 22 '11 at 04:16
  • @BrandonBertelsen - `for(1:as.numeric(test))` doesn't iterate through the number of days in `test`, it starts at Jan 1, 1970, and counts forward the number of days elapsed until the day `test`. – Ken Williams Sep 06 '12 at 21:20
  • @Ken, by setting 1:as.numeric(test) you're telling it to start at 1, hence Jan 1, 1970. – Brandon Bertelsen Sep 07 '12 at 00:36
  • 2
    Right, but that's not often a useful thing to do. 1970 wasn't *that* great a year. ;-) @SFun28 was interested in iterating through the multiple entries of `test`, not from some external start date to `test`. And if `test` has length more than 1, `1:as.numeric(test)` issues a stern warning and discards all but the first element. – Ken Williams Sep 07 '12 at 04:09
0

Not a solution per-se, but a useful trick when you want to check the dates in a loop:

for( i.date in as.character(Sys.Date()) ){ cat(paste("Date:", i.date, "\n")) }

just transform it as character beforehand. Most filters won't bother.

Xavier Prudent
  • 1,570
  • 3
  • 25
  • 54