option to return sample trajectories

[elray1]

My understanding is that currently, forecasters in this package can produce predictive quantiles. However, in many disease forecasting hubs, there is interest in collecting sample trajectories. It could be nice to add options to return those instead of or in addition to quantiles.

[dshemetov]

My apologies, probably my lack of familiarity, but how are sample trajectories defined outside the Monte Carlo context? Our currently-included forecasters fit models to a specific ahead (e.g. 7 days out), without the need to produce the intermediate aheads (e.g. 1 through 6 days out), so there's no notion of a trajectory per se. We have thought about adding in iterative models (ones that produce 1 ahead at a time and feed their own output back in to produce a series of ahead forecasts), but haven't worked out how to integrate that in our framework yet. Maybe the latter is what you have in mind? (Though the forecaster output format for the latter would still be a hub-style prediction card... but I suspect you're not talking just about a format differences.)

[elray1]

Good question -- I had forgotten that your models generally fit to a specific ahead, and you're right -- I had iterative methods in mind. (Is the cdc flatline forecaster using iterative methods?)

[dshemetov]

Ah yes, I think that's the only that works that way.