How to convert time series dates into data frame dates-CodePudding

I have a time series of weekly data, beginning at Jan. 1, 2016. I've tried using the method in this question, but getting dates from 1970.

This is what I'm doing below:

# Creating this df of dates used later on
index.date <- data.frame(start.date=seq(from=as.Date("01/01/2016",format="%m/%d/%Y"),
                                    to=as.Date("10/30/2021",format="%m/%d/%Y"), 
                                    by='week'))

# Create a ts, specifying start date and frequency=52 for weekly
weekly.ts <- ts(rnorm(305,0,1),start=min(index.date$start.date), frequency = 52)

# Look at the min and max dates in the ts
as.Date(as.numeric(time(min(weekly.ts))))
[1] "1970-01-02"
as.Date(as.numeric(time(max(weekly.ts))))
[1] "1970-01-02"

I plan to place the ts into a df with dates shown in a date format with the following:

# Place ts dates and values into a df
output.df <-data.frame(date=as.Date(as.numeric(time(weekly.ts))), 
                   y=as.matrix(weekly.ts))

Is this a matter of me specifying the dates incorrectly in the ts, or am I converting them incorrectly with as.Date(as.numeric(timeweekly.ts))))? I would expect the min date to be Jam. 1, 2016, and the maximum Oct. 29, 2021 (as it is for index.date).

CodePudding user response：

Try this

weekly.ts <- ts(rnorm(305,0,1),
      start=min(index.date$start.date),
      end=max(index.date$start.date), frequency=2)

# look at plot to see if it works
plot(stl(weekly.ts, s.window=2))

# get time
head(as.POSIXlt.Date(time(weekly.ts)))
[1] "2016-01-01 UTC" "2016-01-01 UTC" "2016-01-02 UTC" "2016-01-02 UTC"
[5] "2016-01-03 UTC" "2016-01-03 UTC"

tail(as.POSIXlt.Date(time(weekly.ts)))
[1] "2021-10-26 UTC" "2021-10-27 UTC" "2021-10-27 UTC" "2021-10-28 UTC"
[5] "2021-10-28 UTC" "2021-10-29 UTC"

You get 2 dates because of freqency=2, which is required by decompose or stl for meaningful data.

CodePudding user response：

ts series do not understand Date class but you can encode the dates into numbers and then decode them back. Assuming that you want a series with frequency 52 the first week in 2016 will be represented by 2016, the second by 2016 1/52, ..., the last by 2016 51/52.

For example,

tt <- ts(rnorm(305), start = 2016, freq = 52)

Now decode the dates.

toDate <- function(tt) {
  yr <- as.integer(time(tt))
  week <- as.integer(cycle(tt))  # first week of year is 1, etc.
  as.Date(ISOdate(yr, 1, 1))   7 * (week - 1)
}

data.frame(dates = toDate(tt), series = c(tt))

We can also convert from Date class to year/week number

# input is a Date class object
to_yw <- function(date) {
  yr <- as.numeric(format(date, "%Y"))
  yday <- as.POSIXlt(date)$yday  # jan 1st is 0
  week <- pmin(floor(yday / 7), 51)   1  # 1st week of yr is 1
  yw <- yr   (week - 1) / 52
  list(yw = yw, year = yr, yday = yday, week = week)
}