How to avoid data duplicates when using slider_index() with multiple observations per index value?

user_10264278 0
Viewed 34

I am using slider::slider_index() to set up a rolling time windows over a dataset. The idea is to apply a function to all points belonging to the time window.

Here is a reproducible example to get all individuals (1 individual = 1 row of the dataframe) belonging to 2 consecutive days.

# packages
library(dplyr)
library(lubridate)
library(slider)
# dummy data
company <- tibble(
  sales = c(2, 3, 1, 8, 4, 6, 2),
  index = as.Date("2019-08-29") + c(0, 0, 1, 1, 1, 5, 6),
)
# what I have done so far
roll_index <- slide_index(company, company$index, ~ .x, .before = lubridate::days(1))

Which gives:

> roll_index
[[1]]
# A tibble: 2 x 2
  sales index     
  <dbl> <date>    
1     2 2019-08-29
2     3 2019-08-29

[[2]]
# A tibble: 2 x 2
  sales index     
  <dbl> <date>    
1     2 2019-08-29
2     3 2019-08-29

[[3]]
# A tibble: 5 x 2
  sales index     
  <dbl> <date>    
1     2 2019-08-29
2     3 2019-08-29
3     1 2019-08-30
4     8 2019-08-30
5     4 2019-08-30

[[4]]
# A tibble: 5 x 2
  sales index     
  <dbl> <date>    
1     2 2019-08-29
2     3 2019-08-29
3     1 2019-08-30
4     8 2019-08-30
5     4 2019-08-30

[[5]]
# A tibble: 5 x 2
  sales index     
  <dbl> <date>    
1     2 2019-08-29
2     3 2019-08-29
3     1 2019-08-30
4     8 2019-08-30
5     4 2019-08-30

[[6]]
# A tibble: 1 x 2
  sales index     
  <dbl> <date>    
1     6 2019-09-03

[[7]]
# A tibble: 2 x 2
  sales index     
  <dbl> <date>    
1     6 2019-09-03
2     2 2019-09-04

So far so good. The problem is that it is not exactly what I expected. As you can see, some data are duplicated: roll_index[[3]], roll_index[[4]] and roll_index[[5]] are the same. What I understand is that slider::slide_index() goes one individual after the other even if multiple individuals belong to the same index value.

How can I avoid this data duplicates as they tend to increase quite a lot the size of the ouput when I use my actual data?

r time
1 Answers

What you can do is use the slider::hop() set of functions that form the backbone of the slide functions. In your case, hop_index() does the trick. We just need to manually set the .stop to the unique values of company$index and .start to be the day before.

stops <- unique(company$index)
starts <- stops - lubridate::days(1)

hop_index(
  .x = company,
  .i = company$index,
  .starts = starts,
  .stops = stops,
  .f = ~.x
)
#> [[1]]
#> # A tibble: 2 × 2
#>   sales index     
#>   <dbl> <date>    
#> 1     2 2019-08-29
#> 2     3 2019-08-29
#> 
#> [[2]]
#> # A tibble: 5 × 2
#>   sales index     
#>   <dbl> <date>    
#> 1     2 2019-08-29
#> 2     3 2019-08-29
#> 3     1 2019-08-30
#> 4     8 2019-08-30
#> 5     4 2019-08-30
#> 
#> [[3]]
#> # A tibble: 1 × 2
#>   sales index     
#>   <dbl> <date>    
#> 1     6 2019-09-03
#> 
#> [[4]]
#> # A tibble: 2 × 2
#>   sales index     
#>   <dbl> <date>    
#> 1     6 2019-09-03
#> 2     2 2019-09-04
user_103573830
Related