Last week Paul (@goodvirus) kindly helped me to generate below workflow for ARIMA forecasts which takes care of multiple (potentially infinite) time series predictions. Now I came across next issue: I would like to generate multiple ahead forecasts with rolling origins. It means let’s say I’m interested to see how trained model (which is trained on First Partition and of course parameters would be fixed since then) would predict 4-steps ahead data points in Second Partition. Since the second partition in my data (in below workflow) has 12 data points, it means I need to get 12 4-step ahead forecasts (naturally originated from 12 different origins but all from one model). Hope my explanation was not too vague!
This is a very interesting topic in general and question too in particular. Thanks for posting the workflow. Maybe you could modify your post (while still possible) to add a link to the previous thread where @goodvirus posted this workflow so that we can have all the complementary information. Thanks in advance.