date windows behaving unexpectedly

[danwwilson]

Firstly, let me express my sincere appreciation for this package. It's been super helpful when dealing with rolling measure on date data with gaps.

I came across a behaviour that seemed strange to me and I wanted to check my understanding (and maybe report a bug).

The following reprex shows the issue I've encounted.

  library(data.table)
  dt  <- data.table(
    date = as.Date(c("2025-07-01", "2025-08-01", "2025-09-01")),
    b = c(10, 20, 30)
  )
  
  dt[,
    let(
      rolling_2m= runner(x=b, k = "2 months", idx = date, f = sum, na_pad = TRUE))]
  dt
#          date     b rolling_2m
#        <Date> <num>      <num>
# 1: 2025-07-01    10         NA
# 2: 2025-08-01    20         NA
# 3: 2025-09-01    30         50

I'm curious to understand why the second record doesn't end up with a value in rolling_2m when it's the second month. Any guidance here would be appreciated.

[gogonzo]

Thanks for the kind words and for the clear reprex.

This is expected behaviour — na_pad = TRUE tells runner() to return NA whenever the window extends before the first observation in your data. For row 2 (2025-08-01), the 2-month window starts at 2025-06-01, but your data only begins in July, so the window is considered incomplete.

To get the sum over whatever data falls within the window (even partial windows), use na_pad = FALSE:

dt[, rolling_2m := runner(x = b, k = "2 months", idx = date, f = sum, na_pad = FALSE)]
#          date     b rolling_2m
# 1: 2025-07-01    10         10
# 2: 2025-08-01    20         30
# 3: 2025-09-01    30         50

The naming of na_pad is admittedly not obvious here — we'll look at improving the documentation to make this distinction clearer with a date-based example.