https://github.com/nixtla logo
m

marah othman

09/05/2023, 8:31 AM
I have a question about RNN Models do these models in the training stage predict one by one or all the sequences at once?
c

Cristian (Nixtla)

09/05/2023, 5:35 PM
Hi @marah othman! The step size is one, so they make predictions one by one (not the entire sequence at once). However, each prediction will consist of
h
future values, and those
h
values are produced at once.
So for example if
h=3
, at timestamp
t
it will first produce the forecast for
t+1
,
t+2
, and
t+3
simultaneously. Then it will move once, to
t+1
, and forecast for
t+2
,
t+3
, and
t+4
. And so on
a

Antoine SCHWARTZ -CROIX-

09/06/2023, 10:02 AM
Hi @Cristian (Nixtla), I know we've already had a quick chat about this, but are you planning in your short-term roadmap to add the option allowing you to dynamically recalculate lag features on the target at each step, for all recursive approaches (like in MLForecast) ?
m

marah othman

09/08/2023, 11:35 AM
Hi @Cristian (Nixtla) ,I mean by all the sequences at once I meant the whole horizon So for example if
h=3
, at timestamp
t
it will first produce the forecast for
t+1
,
t+2
, and
t+3
simultaneously. Then it will move once, to
t+1
, and forecast for
t+2
,
t+3
, and
t+4
. And so on like what you explain here
I mean (and those
h
values are produced at once.) that is what I am asking if at once or one by one.
2 Views