Slider (slide_period) Moving Average Calculation Wrong - r

Question

I'm trying to use Slider to compute moving averages over some time series data. The data has day resolution (one observation per day). For each observation I want to compute the average daily value over the last 7 days.
The problem is my code is ignoring the missing observations with implied values of zero. So if my period is 7 days, and during some 7 day window there are only 2 observations, it's summing them and dividing by 2, whereas I'm looking to sum and divide by 7 to get the average per day.
In the code below you'll see that the second row (2023-02-03) is computing the average by dividing by 2 (the number of observations), rather than by dividing by 4 (the number of days in the period 2023-01-31 to 2023-02-03).
Is there a good way to achieve the desired result, or do I just need to replace the mean calculation with sum() / 7?
I had originally backfilled the missing observations which worked, but the data is relatively large and quite sparse and doing so massively increased the runtime (from ~8 seconds to ~100).
library(tidyverse)
library(slider)
data <- data.frame(
date = Sys.Date() - c(0, 1, 2, 3, 4, 6, 7, 8, 9, 10, 13),
val = c(0, 0, 2, 1, 0, 10, 0, 1, 1, 6, 1)
)
print(as_tibble(data))
summary <- function(data) {
summarise(data,
moving_total = sum(val),
moving_avg = mean(val, na.rm = FALSE),
num_observations = n()
)
}
res <- data %>%
arrange(date) %>%
mutate(
weekly = slide_period_dfr(
.x = pick(everything()),
.i = date,
.period = "day",
.f = summary,
.before = 6,
.complete = FALSE
)
)
print(as_tibble(res))
# A tibble: 11 x 2
date val
<date> <dbl>
1 2023-02-13 0
2 2023-02-12 0
3 2023-02-11 2
4 2023-02-10 1
5 2023-02-09 0
6 2023-02-07 10
7 2023-02-06 0
8 2023-02-05 1
9 2023-02-04 1
10 2023-02-03 6
11 2023-01-31 1
# A tibble: 11 x 3
date val weekly$moving_total $moving_avg $num_observations
<date> <dbl> <dbl> <dbl> <int>
1 2023-01-31 1 1 1 1
2 2023-02-03 6 7 3.5 2
3 2023-02-04 1 8 2.67 3
4 2023-02-05 1 9 2.25 4
5 2023-02-06 0 9 1.8 5
6 2023-02-07 10 18 3.6 5
7 2023-02-09 0 18 3 6
8 2023-02-10 1 13 2.17 6
9 2023-02-11 2 14 2.33 6
10 2023-02-12 0 13 2.17 6
11 2023-02-13 0 13 2.17 6
Just a note on the implementation. In the real world the moving averages are being computed over groups, hence the use of pick(everything()) above. Don't think it's necessary for the toy example, but I leave it in just in case it influences the answer.
Thanks

Answer 1

A straightforward solution is to supply zero values for the absent dates.
library(tidyverse)
library(slider)
(data_ <- tibble(
date = Sys.Date() - c(0, 1, 2, 3, 4, 6, 7, 8, 9, 10, 13),
val = c(0, 0, 2, 1, 0, 10, 0, 1, 1, 6, 1)
))
summary <- function(.data) {
summarise(.data,
moving_total = sum(val),
moving_avg = mean(val, na.rm = FALSE),
num_observations = n()
)
}
(full_dates <- tibble(date=
seq(min(data_$date),
max(data_$date),
by = "1 day"
)))
(fdj <- left_join(
full_dates,
data_
) |> mutate(val = if_else(
is.na(val), 0, val
)))
(res <-
mutate(fdj,
weekly = slide_period_dfr(
.x = pick(everything()),
.i = date,
.period = "day",
.f = summary,
.before = 6,
.complete = FALSE
)
))