Figure 2 rough draft components

[kaitejohnson]

The point of this figure is to show an example comparing the forecasts of the wastewater model and the hospital admissions model for a few states (can decide on the states later), paired with the fits/forecasts to a subset of the wastewater data for that state.

This PR creates (separately) the ggplot components for the example hospital admissions forecasts colored by model type + the fits and forecasts to the wastewater concentrations for those example states.

See private repo for produced figures.

Note: The goal here is just to have the structure in place to generate the figures. All formatting decisions will be made in a second round of revisions. Am still using the {targets} pipeline for creating all of these, which enables caching and allows us to pass the post-processed objects/ mix analysis in with figure production when needed.

[dylanhmorris]

Looks largely good. A couple minor comments.

[kaitejohnson]

Converting back to a draft because I realized I forgot to make the second half of what was planned for Fig 2 (the forecast comparison plots by location and horizon, with CRPS underlaid)

[kaitejohnson]

Ok @dylanhmorris this is ready for review. Let me know if you'd prefer I just put all the figures in one PR, I figured breaking it up would be slightly easier. All the drafts are in the link to the cdcent repo issue.

In order to make the completed planned figure 2, we'd have to repeat a bunch of generations of the subpanels for the different locations, I left that out of the targets pipeline for now.

[dylanhmorris]

A couple minor but key questions @kaitejohnson

[kaitejohnson]

@dylanhmorris Let me know what you think of this. I added the additional horizon_weeks and horizon columns upstream when creating the long form of the quantiles and the scores, and then pass them in to the figure functions. I think this will avoid redundancy.

I still have an open question about what horizon_week to assigned the forecast date (see my teams post). Once we've reached a consensus will edit that.

[kaitejohnson]

Update, going to switch to creating horizon_days folder and then creating categorical horizon columns from that. Going to lump the forecast date into the nowcast, since in theory, you could have partially observed data on the day you are making a forecast

[kaitejohnson]

Ok this is ready for re-review, let me know if you want me to make them into one function though @dylanhmorris

[dylanhmorris]

LGTM. One documentation wording suggestion.