struggling to regularize a time series using tsibble

Question

I'm struggling to regularize a time series using the tsibble package. The documentation indicates that this can be done using index_by() + summarise(), but I'm clearly missing some details. Here's what I've tried:

library(tidyverse)
library(lubridate)
library(tsibble)

# example data set
date <- ymd(c("1976-05-18", "1976-05-19", "1976-05-24", "1976-06-01"))
fish <- c(203, 282, 301, 89)
volume <- c(210749, 287555, 378965, 308935)
n <- c(5, 7, 10, 8)
tbl <- tibble(date, fish, volume, n)
tsbl <- tsibble(tbl, index = date, regular = FALSE)
  
# regularize the tsibble (ie time series)
tsbl %>% 
  index_by(date, unit = "day") %>% # unit value "day" is intuitive but incorrect?
  mutate(week = isoweek(date)) %>% # add (numeric) week column
  summarise(date = date,
            fish = sum(fish),
            volume = sum(volume),
            n = sum(n), 
            cpue = fish/volume) # calculate catch per unit effort

TIA!

Just set `regular = TRUE` in the `tsibble` creation. The summarise isn't doing anything -- did you mean to group by week? — Rob Hyndman, May 19 '22 at 22:00
Setting ```regular = TRUE``` didn't work for me. I get a tsibble of 4 rows with the same irregular interval btwn rows. ```tsbl <- tsibble(tbl, index = date, regular = TRUE)``` I was able to regularize the example using ```complete(tbl, date = seq.Date(min(date), max(date), by = "day"))```. What am I missing? — Peter Nelson, May 19 '22 at 23:35

score 3 · Accepted Answer · answered May 20 '22 at 02:56

With so little information provided about what you are actually trying to do, I will have to guess.

Perhaps you want daily data with each day explicitly included. In that case, do this:

library(tidyverse)
library(lubridate)
library(tsibble)

# example data set
date <- ymd(c("1976-05-18", "1976-05-19", "1976-05-24", "1976-06-01"))
fish <- c(203, 282, 301, 89)
volume <- c(210749, 287555, 378965, 308935)
n <- c(5, 7, 10, 8)
tbl <- tibble(date, fish, volume, n)
tsbl <- tsibble(tbl, index = date, regular = TRUE) %>%
  fill_gaps()
tsbl
#> # A tsibble: 15 x 4 [1D]
#>    date        fish volume     n
#>    <date>     <dbl>  <dbl> <dbl>
#>  1 1976-05-18   203 210749     5
#>  2 1976-05-19   282 287555     7
#>  3 1976-05-20    NA     NA    NA
#>  4 1976-05-21    NA     NA    NA
#>  5 1976-05-22    NA     NA    NA
#>  6 1976-05-23    NA     NA    NA
#>  7 1976-05-24   301 378965    10
#>  8 1976-05-25    NA     NA    NA
#>  9 1976-05-26    NA     NA    NA
#> 10 1976-05-27    NA     NA    NA
#> 11 1976-05-28    NA     NA    NA
#> 12 1976-05-29    NA     NA    NA
#> 13 1976-05-30    NA     NA    NA
#> 14 1976-05-31    NA     NA    NA
#> 15 1976-06-01    89 308935     8

^{Created on 2022-05-20 by the reprex package (v2.0.1)}

I'm not sure what you are trying to achieve with the summarize, but perhaps you want to create weekly data from these daily data. In that case, do this:

tsbl %>% 
  mutate(week = isoweek(date)) %>% # add (numeric) week column
  index_by(week) %>%
  summarise(fish = sum(fish, na.rm=TRUE),
            volume = sum(volume, na.rm=TRUE),
            n = sum(n, na.rm=TRUE), 
            cpue = fish/volume) # calculate catch per unit effort
#> # A tsibble: 3 x 5 [1]
#>    week  fish volume     n     cpue
#>   <dbl> <dbl>  <dbl> <dbl>    <dbl>
#> 1    21   485 498304    12 0.000973
#> 2    22   301 378965    10 0.000794
#> 3    23    89 308935     8 0.000288

^{Created on 2022-05-20 by the reprex package (v2.0.1)}

Thanks! Sorry for the lack of detail re my ```summarise()``` efforts. Your answers are super helpful (you guessed correctly). — Peter Nelson, May 20 '22 at 06:16
@PeterNelson, please accept the answer. It will then be marked as solved. — phiver, May 20 '22 at 06:59

struggling to regularize a time series using tsibble

1 Answers1