As mentioned in the title, would that count in the allocated compute budget?
I feel like it should not count because if it does, it would have a big impact on the first week’s compute budget. This means that we would have to add some logic of the sort:
-if week==1, then do not do extra train, else, do it.
Adding this logic is possible, but I do not know if this is really necessary.
Thanks for you clarifications @enzo
Actually, now that I think of it…
Would that simply mean that the period starts at time_id = 1?
Runs will be run every weeks with 16h of maximum runtime.
The next week will be the same, the 16h will be reset.
You should not have to worry about the timings since you will only be run on 3 dates instead of 30.
Sorry I do not understand what you mean…
Are you saying that the “force first retrain” is essentially its own week (i.e., counts as a week with 16h compute and after that the counter gets reset)?
Each week will have his own dedicated 16h.
The force first train choice dictate if for the very first Run of the out of sample week, the
train should be called even though you already submitted a
Okay, I understand I think.
So it would use up computing time allocated to the first week (i.e., first three dates):
- 0-298, predict on 300
- 0-299, predict on 301
- 0-300, predict on 302
thanks for the quick replies