0

I'm working with time series data and I'm curious what does week2, week3 and week4 refer to when using tk_augment_timeseries_signature

I'm working with the weekly data (walmart_sales_weekly) from the timetk package:

data_tbl <- timetk::walmart_sales_weekly %>%
  select(id, Date, Weekly_Sales) %>%
  mutate(id = as.factor(id)) %>% 
  set_names(c("id", "date", "value")) %>% 
  filter(id == "1_95") %>%
  mutate(value = log1p(x = value)) %>%
  mutate(value = standardize_vec(value)) %>%
  tk_augment_timeseries_signature(.date_var = date) %>%
  select(-diff, -matches("(.xts$)|(.iso$)|(hour)|(minute)|(second)|(am.pm)|(wday)|(day)|(mweek)"))

The help documentation i.e. ?tk_augment_timeseries_signature makes no reference to these features

TheGoat
  • 2,587
  • 3
  • 25
  • 58

1 Answers1

1

Based on the vignette in their github page, they are referring to the different frequency based upon the week time feature (week2 for bi-weekly frequency, week3 for tri-weekly frequency). They are created through the use of modulus.

John Manacup
  • 314
  • 1
  • 4
  • Thanks John, I changed the original question to specify the help docs, good find on the github page. Unfortunately for me I am still not following, the frequency and modulus logic. The date '2010-02-12' have 1,1 & 3 populated for week2, 3 & 4 respectively, can you share how these figures are calculated? Thanks – TheGoat Oct 25 '22 at 21:03