Lesson3-rossman: Is "Time until next event" generally possible?

In this lesson, in the section “Durations”, there’s an item called “Time until next event”. Is this generally feasible? If the event is random or unknown ahead of time, how would we know this at test/deploy time? And if the event happened at regularly interval, then wouldn’t this be completely derivable from “Time since last event” and offers no additional useful information.

I am not sure the use of “Time until next event” is very specific to this competition and its dataset. For time series, you probably don’t have information about a future stochastic event.

Can someone offer additional explanation? Thanks.

I think ‘event’ referred to a store promotion, which would be scheduled by store management.

Thanks. I read the code further down. It seemed this get_elapsed(…) is responsible for calculating the “time since/until next event” has to do with

  1. school holiday
  2. state holiday
  3. promotion event

Now, i can see it may make sense for the holidays. They are not regular but they are totally “predictable” events being fixed. Also, on 2nd thought, i think promotion event may also make sense, since internal management will know this. But i think it is sort of a weaker indicator since customers may not know when the next promotion event going to be.

Probably weaker in some cases. Consumers might expect a ‘back to school sale’ promotion in late summer and hold off on their shopping in anticipation, but other promotions might be less predictable.