Get graph of the weekly season for a time series with daily frequency

0
votes

How can I decompose a time series of daily frequency in R? I have the number of visitors to a website each day over the course of a year. I want to show a graph of the weekly season.

my_data = read.csv("time series test.csv", header = TRUE)

my_zoo_ts = read.zoo("time series test.csv", sep = ",", format="%m/%d/%y")

stl(my_zoo_ts)
# Error in stl(coerced_ts) : 
#   series is not periodic or has less than two periods

I was hoping to use STL, but apparently STL can't be used with daily data (from another S.O. question).

Here is the head of my data.

head(my_data)
      V1  V2
1 1/1/14 123
2 1/2/14 128
3 1/3/14 129
4 1/4/14 130
5 1/5/14 137
6 1/6/14 141
2
rtime-serieszoo
It can be used for daily data. But with more than two years of data. Anyway, what do you want to decompose? – user3710546
stl cannot be used with "Date" class data. The data must be regularly spaced and one period must have the duration of 1. – G. Grothendieck
@Pascal - I want to decompose a time series where it's the number of unique visitors to a website each day over 1 year. The periods would be weeks (since people visit certain websites more on certain days of the week). Is there any other function I can use? – Don P
@G.Grothendieck - thanks I can create a series with zoothat is just a single vector regularly spaced. But I can't figure out how to specify the period to be 1 week. – Don P
I still don't understand. Do you want to aggregate daily to weekly data? – user3710546

2 Answers

0
votes

You should either aggregate the week or aggregate over the day-of-the-week.

Then you can plot the aggregates.

If you need further details, can you explain what you want to see?

0
votes
z <- as.Date("2014-1-1") + 0:364
visits <- sample(seq(from = 50, to = 100, by = 1), size = 365, replace = TRUE)
z.day <- zoo(visits, z)
week <- as.numeric(format(time(z.day), "%W"))
z.week <- aggregate(z.day, week, sum)

Then plot z.week