Feeding an hourly zoo time-series into function stl()

Question

Before you ask, yes I need to show this much data. stl() requires two periods of data. In this case, one period is 24 values, so stl() wants at least 48 values.

Also, from stl() help:

"....This should be an object of class "ts" with a frequency greater than one...."

I'm upgrading some old calcs so that my data is in the zoo format. So far, I've upgraded monthly and daily data without any noticeable problems, although there have been stl() speed issues. I'm now down to hourly data.

When you run the code, the "First Try" and "Second Try" fail while the "Third Try" works. Why? What really obvious thing have I missed? From the "notice what happens next" section, is it stl()'s internal conversion from a zoo series to a ts series that is causing this problem?

library(zoo)   

texinp <- "
Date,Demand
2009-01-01 01:00:00,28099.1458685981
2009-01-01 02:00:00,27818.8853634983
2009-01-01 03:00:00,27542.1297303602
2009-01-01 04:00:00,27481.2230897352
2009-01-01 05:00:00,27863.5201611328
2009-01-01 06:00:00,28716.5855560981
2009-01-01 07:00:00,29542.9219102648
2009-01-01 08:00:00,29894.8927398003
2009-01-01 09:00:00,29984.6039306641
2009-01-01 10:00:00,30065.9059781901
2009-01-01 11:00:00,29885.6435823568
2009-01-01 12:00:00,29348.3372466363
2009-01-01 13:00:00,28547.9348426649
2009-01-01 14:00:00,27601.1429031033
2009-01-01 15:00:00,26784.0209678819
2009-01-01 16:00:00,26269.0193386502
2009-01-01 17:00:00,26196.6366764323
2009-01-01 18:00:00,27527.2755148655
2009-01-01 19:00:00,29723.9232834201
2009-01-01 20:00:00,29594.3683930122
2009-01-01 21:00:00,29089.5591764323
2009-01-01 22:00:00,28287.9040272352
2009-01-01 23:00:00,26863.3280593533
2009-01-02 00:00:00,25166.2158816189
2009-01-02 01:00:00,23708.111414388
2009-01-02 02:00:00,22905.427265625
2009-01-02 03:00:00,22517.7926079644
2009-01-02 04:00:00,22519.1570795356
2009-01-02 05:00:00,23065.7224479167
2009-01-02 06:00:00,24452.6789577908
2009-01-02 07:00:00,26450.5363346354
2009-01-02 08:00:00,27815.8465516493
2009-01-02 09:00:00,28402.8746001519
2009-01-02 10:00:00,29007.9232600911
2009-01-02 11:00:00,29333.0119395616
2009-01-02 12:00:00,29378.4373334418
2009-01-02 13:00:00,29145.4215820312
2009-01-02 14:00:00,29069.2706928168
2009-01-02 15:00:00,28900.4390755208
2009-01-02 16:00:00,28724.9172607422
2009-01-02 17:00:00,28523.1717095269
2009-01-02 18:00:00,29480.8798573134
2009-01-02 19:00:00,31168.033992513
2009-01-02 20:00:00,30736.5264789497
2009-01-02 21:00:00,29963.2411859809
2009-01-02 22:00:00,28915.5116943359
2009-01-02 23:00:00,27547.5425157335
2009-01-03 00:00:00,25756.0379166667
2009-01-03 01:00:00,24079.5182259115
2009-01-03 02:00:00,22927.8879052734"


tem <- read.zoo(textConnection(texinp), header = TRUE, sep = ",", tz = "", regular = TRUE)

#First try
frequency(tem)
tem.stl <- stl(tem, s.window = "periodic") #error, not the frequency it was expecting?

#Second try
frequency(tem) <- 24
tem.stl <- stl(tem, s.window = "periodic") #don't worry, it's not hung....
#r calculates for about a minute and comes back with an error

#Third try
temcor <- ts(coredata(tem), freq = 24)
temcor.stl <- stl(temcor, s.window = "periodic") #works fine
plot(temcor.stl)

#Also, notice what happens next
junk <- as.ts(tem) #again, it's not hung, it just takes a while
str(junk)
frequency(junk)
junk #hello

Answer 1

Use a time representation for which is a full cycle is 1 so for hourly data with a frequency of 24 a unit of time should correspond to a day. chron works that way so:

library(zoo)
library(chron)

z <- read.zoo(text = texinp, header = TRUE, sep = ",", FUN = as.chron)
stl(z, "per")