Subset xts time-series object in R

Question

I have time-series xts object for certain months like this

library(xts)
  seq<- seq(as.POSIXct("2015-09-01"),as.POSIXct("2015-09-04"), by = "30 mins")
  ob<- xts(data.frame(power=1:(length(seq))),seq)

Now, corresponding to each observation (say A ) I want to calculate mean of the last two hours observations. Therefore, corresponding to each observation ( A ) I need to calculate index of the observation happened before two hours to A , say it is B . Then I can calculate mean of the observations between A and B . Accordingly,

i=10 # dummy
ind_cur<- index(ob[i,]) # index of current observation
ind_back <- ind_cur - 3600 * 2 # index of 2 hours back observation

With these indices, I am subsetting ob as

 ob['ind_cur/ind_back']

It results in following error:

Error in if (length(c(year, month, day, hour, min, sec)) == 6 && c(year,  : 
  missing value where TRUE/FALSE needed
In addition: Warning messages:
1: In as_numeric(YYYY) : NAs introduced by coercion
2: In as_numeric(MM) : NAs introduced by coercion
3: In as_numeric(DD) : NAs introduced by coercion
4: In as_numeric(YYYY) : NAs introduced by coercion
5: In as_numeric(MM) : NAs introduced by coercion
6: In as_numeric(DD) : NAs introduced by coercion

Can anyone help me to subset ob ! Found a related question at the link , but not enough to solve this issue.

Update Expected output shown as

2015-09-01 00:00:00     1   NA # as I don't have previous data
2015-09-01 00:30:00     2   NA
2015-09-01 01:00:00     3   NA
2015-09-01 01:30:00     4   NA
2015-09-01 02:00:00     5   10/4 # mean of prevous 4 observations (last two hours)
2015-09-01 02:30:00     6   14/4  
2015-09-01 03:00:00     7   18/4

Answer 1

This is a difficult problem to solve generally, so you need to roll your own solution. The easiest is to use window to subset by overlapping 2-hour intervals.

# initialize a result object
ob2 <- ob * NA_real_
# loop over all rows and calculate 2-hour mean
for(i in 2:nrow(ob)) {
  ix <- index(ob)[i]
  ob2[i] <- mean(window(ob, start=ix-3600*2, end=ix))
}
# set incomplete 2-hour intervals to NA
is.na(ob2) <- which(index(ob2) < start(ob2)+3600*2)

Answer 2

We could use rollapply() package in combination with lag() to offset the resulting rolling mean by one row.

rollapply(lag(ob), 4, mean)
#                    power
#2015-09-01 00:00:00    NA
#2015-09-01 00:30:00    NA
#2015-09-01 01:00:00    NA
#2015-09-01 01:30:00    NA
#2015-09-01 02:00:00   2.5
#2015-09-01 02:30:00   3.5
#2015-09-01 03:00:00   4.5

# Or if you want it as new variable in your xts object
ob$mean <- rollapply(lag(ob),4,mean)

Answer 3

Based on the update to the question "Expected output" and the comment by RS:

library(TTR)
head(SMA(ob$power, 4))  # 2 hour moving average

result

                    SMA
2015-09-01 00:00:00  NA
2015-09-01 00:30:00  NA
2015-09-01 01:00:00  NA
2015-09-01 01:30:00 2.5
2015-09-01 02:00:00 3.5
2015-09-01 02:30:00 4.5

This assumes the 30 minute interval stated in question.

To look more exactly like Expected Output:

lag(head(SMA(ob$power, 4),7))

                    SMA
2015-09-01 00:00:00  NA
2015-09-01 00:30:00  NA
2015-09-01 01:00:00  NA
2015-09-01 01:30:00  NA
2015-09-01 02:00:00 2.5
2015-09-01 02:30:00 3.5
2015-09-01 03:00:00 4.5

Answer 4

Package data.table offers a rolling function, useful for single- as well as multiple time series:

head(

    as.data.table(ob)[, roll_power := frollmean(power, 4, align = 'right')]
)

# at the end of a 4 1/2 hour lag

                 index power roll_power
1: 2015-09-01 00:00:00     1         NA
2: 2015-09-01 00:30:00     2         NA
3: 2015-09-01 01:00:00     3         NA
4: 2015-09-01 01:30:00     4        2.5 # the rolling mean covers this, and preceding rows
5: 2015-09-01 02:00:00     5        3.5
6: 2015-09-01 02:30:00     6        4.5