CODEWITHSUNDEEP
rtime-series
bill_080
bill_080

Reputation: 4760

Calculating Start for stl()

I'm reading weekly data from a .csv data file. A sample of the data is: 

Date,Demand    
"Feb 08, 1991",6621    
"Feb 15, 1991",6433    
"Feb 22, 1991",6582   
"Mar 01, 1991",7224   
"Mar 08, 1991",6875   
"Mar 15, 1991",6947   
"Mar 22, 1991",7328   
"Mar 29, 1991",6777   
"Apr 05, 1991",7503
.....

My code is:

> temp<-read.table(file="E:\\Data\\Demand_00.csv",header=TRUE, sep=",")
> stadat<-strptime(as.character(temp[,1]),"%b %d, %Y")[1]
> statim<-as.numeric(strftime(stadat,"%Y"))+(as.numeric(strftime(stadat,"%j"))/366)
> temdat<-ts(temp[,2],start=statim,frequency=52)
> plot(temp2<- stl(log(temdat), "per"))

My question is: Is there a better/cleaner way to build statim (the start required in the above ts object)? Notice that this is weekly data that may or may not start at the first week of the year.

Thanks,
Bill

Upvotes: 0

Views: 296

Answers (1)

G. Grothendieck
G. Grothendieck

Reputation: 269481

You could use the zoo package to simplify this:

File <- E:\\Data\\Demand_00.csv"

library(zoo)
fmt <- "%b %d, %Y"

year.jul <- function(x) as.numeric(format(x, "%Y")) + 
    as.numeric(format(x, "%j"))  / 366
z0 <- read.zoo(File, header = TRUE, sep = ",", FUN = as.Date, format = fmt,
    FUN2 = year.jul)
ts(z0, start = start(z0), frequency = 52)

On the other hand rather than forcing it into 366 days you might want to use cal.yr in the Epi package:

library(Epi)
z2 <- read.zoo(File, header = TRUE, sep = ",", FUN = cal.yr, format = fmt)
as.ts(z2)

Upvotes: 3

Related Questions