[R] POSIXct to ts

Fri Aug 10 09:02:49 CEST 2012

Hi Mary Ann,

I'm afraid I'm not really qualified to answer, though someone else on
this list might be: I might suggest you ask on CrossValidated
(stats.stackexchange.com) where a bunch of nice folks who know far
more than me about these matters answer statistical questions. In
particular, I know Prof Hyndman is seen there on occasion and he could
certainly give you an answer.

Cheers,
Michael

On Thu, Aug 9, 2012 at 11:02 PM, Mary Ann Middleton <maberg at sfu.ca> wrote:
> Hi Michael,
>
>
>
> Thank you so much for your email.  That really helped, and with a frequency
> of 24, I finally have some figures I can work with!!
>
>
>
> I have a follow up question.  Do you have any resources you would recommend
> that would explain the difference between stl() and decompose()?  I have
> been reading the help files, and googling, but I am not finding the right
> resources.
>
>
>
> There are some notable difference between the "random" and the "remainder"
> in each and I am unsure which is correct for my purposes.
>
>
>
> At this time, I can use the "random" calculation from decompose() to
> generate an acf of the data after the seasonal patterns and trends are
> stripped
>
> (acf(na.omit(x.ts.decomp$random)),
>
> which is ultimately what I need, however, that isn't solid justification for
> choosing that calculation.
>
>
>
> Any pointers appreciated.
>
>
>
> Cheers,
>
> Mary Ann
>
> ________________________________
>
> From: "R. Michael Weylandt" <michael.weylandt at gmail.com>
> To: "Mary Ann Middleton" <maberg at sfu.ca>
> Cc: r-help at r-project.org
> Sent: Thursday, August 9, 2012 2:51:16 PM
> Subject: Re: [R] POSIXct to ts
>
> On Thu, Aug 9, 2012 at 3:30 PM, Mary Ann Middleton <maberg at sfu.ca> wrote:
>>
>> Hi,
>>
>> I have a dataframe (try.1) with  date/time and temperature columns, and
>> the date/time is in POSIXct fomat. Sample included below.
>>
>> I would like to to try decompose () or stl() to look at the trends and
>> seasonality in my data, eventually so that  I can  look at autocorrelation.
>> The series is 3 years of water temperature with clearly visible seasonal
>> periods.
>>
>> Right now, if I try decompose, I get the following error, w hich I beleive
>> is because I haven't correctly defined a time series.
>> "Error in decompose(try.1) : time series has no or less than 2 periods"
>>
>> I am stuck trying to go from POSIXct to as.ts
>> Any suggestions on how to tackle this would be greatly appreciated.
>>
>> Sincerely,
>> Mary Ann
>>
>> A sample of the data looks like this:
>> 'data.frame':   26925 obs. of  2 variables:
>>  $ date      : POSIXct, format: "2008-07-11 21:00:00" "2008-07-11
>> 22:00:00" ...
>>  $ DL_1297699: num  15.3 15.1 14.9 14.6 14.1 ...   date DL_1297699 1
>> 2008-07-11 21:00:00     15.318
>> 2       2008-07-11 22:00:00     15.127
>> 3       2008-07-11 23:00:00     14.888
>> 4       2008-07-12 00:00:00     14.553
>> 5       2008-07-12 01:00:00     14.146
>> 6       2008-07-12 02:00:00     13.738
>> 7       2008-07-12 03:00:00     13.401
>> 8       2008-07-12 04:00:00     13.088
>> 9       2008-07-12 05:00:00     12.823
>> 10      2008-07-12 06:00:00     12.630 and the dput(head(x,50) gives this
>> output: structure(list(date = structure(c(1215810000, 1215813600,
>> 1215817200,
>> 1215820800, 1215824400, 1215828000, 1215831600, 1215835200, 1215838800,
>> 1215842400, 1215846000, 1215849600, 1215853200, 1215856800, 1215860400,
>> 1215864000, 1215867600, 1215871200, 1215874800, 1215878400, 1215882000,
>> 1215885600, 1215889200, 1215892800, 1215896400, 1215900000, 1215903600,
>> 1215907200, 1215910800, 1215914400, 1215918000, 1215921600, 1215925200,
>> 1215928800, 1215932400, 1215936000, 1215939600, 1215943200, 1215946800,
>> 1215950400, 1215954000, 1215957600, 1215961200, 1215964800, 1215968400,
>> 1215972000, 1215975600, 1215979200, 1215982800, 1215986400), class =
>> c("POSIXt",
>> "POSIXct"), tzone = "UTC"), DL_1297699 = c(15.318, 15.127, 14.888,
>> 14.553, 14.146, 13.738, 13.401, 13.088, 12.823, 12.63, 12.461,
>> 12.413, 12.461, 12.703, 13.04, 13.497, 14.026, 14.553, 15.031,
>> 15.366, 15.7, 15.819, 15.819, 15.7, 15.605, 15.461, 15.247, 14.984,
>> 14.673, 14.337, 14.002, 13.666, 13.377, 13.137, 12.944, 12.823,
>> 12.847, 13.016, 13.329, 13.762, 14.242, 14.697, 15.175, 15.581,
>> 15.891, 16.034, 16.034, 15.939, 15.772, 15.581)), .Names = c("date",
>> "DL_1297699"), row.names = c(NA, 50L), class = "data.frame")
>>
>
> Thank you for the dput()-ed data!
>
> The "time series" object that stl() and decompose() expect doesn't
> have time stamps -- rather it has a "start" and "end" marker as well
> as a frequency. [For more details, see ?tsp]
>
> With your described data, I'd imagine you'd have start = 2008 and
> frequency = 365*24 (if you have hourly data and an underlying yearly
> periodicity) but to work with the data you gave, lets suppose 12 hours
> is a cycle. Note you don't have to give end because that's figured out
> automatically from frequency and start.
>
> x.ts <- ts(x[,2], start = 1, frequency = 12)
>
> then I can
>
> stl(x, "per")
> decompose(x)
>
> as desired.
>
> Hope that helps,
> Michael
>