Single function pipeline

[stephenturner]

Let's write a single function that strings together everything needed for the forecast we want to produce. For now this will be usa, 1-4 wk ahead ideaths icases cdeaths. Much will be hard-coded and will need to change, and will be obviously open to change as we develop.

[stephenturner]

This needs to be reopened after looking into #26. The first step in this pipeline gets data, then begins modeling. There are arguments that let you pass source and granularity to the get_ functions, but there's a built-in assumption that you have US-level data -- if you pass granularity=state, you'll now have a state column. The join will need to be done by by = c("epiyear", "epiweek", "state") (similar for county), and I imagine everything else may fail downstream.

Options/thoughts:

• quick fix: call this forecast_pipeline_usa, remove the arguments for granularity, and close this issue again for now
• have the pipeline function take data that's already been joined. this might work, but as written, would only work if there was a single location.
• burn it down and start over after fixing issues that #26 is going to reopen in #6 and elsewhere

[vpnagraj]

@stephenturner i might be oversimplifying ... but what if in the get_* functions we changed the name of the column that contains state (our county) to be "location" ... and added "location" as "US" for the granularity = "national"

i think that would make the whole pipeline more flexible. might even be able to do the forecasting with a group_by(location) even for national forecasts ... i think?

[vpnagraj]

so after thinking through this some more ... if we're adding other models / granularity this pipeline function could get unwieldy and not altogether that useful.

assigning this issue to myself to remove the function from the package. in its place i'll put together a pipeline script that will generate, prep, and validate a submission that (for now) includes national and state level TS