Rolling windows with passing threshold in R-CodePudding

I have dataframe like this:

day         device      delta
2021-01-01  A           1.3
2021-01-02  A           2.56
2021-01-03  A           -1.72
2021-01-04  A           4.09
2021-01-05  A           6.99
2021-01-06  A           3.23
2021-01-07  A           1.12
2021-01-08  A           4.56
2021-01-09  A           1.09
2021-01-10  A           2.14
2021-01-01  B           -0.76
2021-01-02  B           -1.78
2021-01-03  B           1.54
2021-01-04  B           1.93
2021-01-05  B           5.98
2021-01-06  B           2.49
2021-01-07  B           8.92
2021-01-08  B           8.08
2021-01-09  B           3.12
2021-01-10  B           1.88

I would like to know whether there is an efficient way of finding out how many 7 days (rows) windows for any of the devices include an increase in the delta value from the first row in the window by more than 5 points?

I would also need to list the first delta value for each window and the first increased value which passes the 5 point threshold.

The desired result would look like this:

window_start    window_end    device    delta_first    delta_first_threshold_pass   
2021-01-01      2021-01-07    A         1.3            6.99
2021-01-03      2021-01-09    A         -1.72          6.99
2021-01-01      2021-01-07    B         -0.76          5.98
2021-01-02      2021-01-08    B         -1.78          5.98
2021-01-03      2021-01-09    B         1.54           8.92
2021-01-04      2021-01-10    B         1.93           8.92

Would something like this be possible? Thank you for any suggestions!

CodePudding user response：

Perhaps you can use rollapplyr from zoo package?

After grouping by device, you can determine the max delta for a 5 day/row window. With the align of "left" the index is left-aligned (includes subsequent days after index date).

Then, you can filter rows where the difference between max value and delta is greater than your threshold of 5.

library(zoo)
library(tidyverse)

df %>%
  group_by(device) %>%
  mutate(max = rollapplyr(delta, width = 5, FUN = max, align = "left", partial = T)) %>%
  filter(max - delta > 5)

Output

  day        device delta   max
  <chr>      <chr>  <dbl> <dbl>
1 2021-01-01 A       1.3   6.99
2 2021-01-03 A      -1.72  6.99
3 2021-01-01 B      -0.76  5.98
4 2021-01-02 B      -1.78  5.98
5 2021-01-03 B       1.54  8.92
6 2021-01-04 B       1.93  8.92
7 2021-01-06 B       2.49  8.92