Forecasting Covid-19 in the US

2020-12-17

Code refactor and review.
Timeseries updated to be daily.
Reimplementation of ensembling and extension to a grid of options.
Cumulative forecast now only calculated at forecast submission.
All previous submissions weeks back cast (excluding the 2020-12-14 and submissions/submitted files) using the standard set of models and an ensemble grid generated for each forecast date.

General tidying:
- get-us-data: added function calls on dplyr
visualisation folder:
Submission plotting:
- Renamed argument for clarity: cutoff is now state_min_cutoff (for selecting states with minimum last week deaths)
- Added back in observed_weeks argument to trim the plot to the last 8 weeks data
Moved raw-data plotting into out-of-date file (untouched otherwise) as not used in main updating scripts
rt-forecast folder:
- rt-forecast/out-of-date has files for raw rt forecast on 2020-06-15 (which saved in a separate folder but are otherwise as expected)
- rt-forecast/submission-files/dated are uniformly formatted properly (ie new formatting code run on forecasts made before the formatting clean-up: 2020-06-15 and 2020-06-22)
- Removed downloading data for each state when running format-rt-fn
- For last week's submission, the up-to-date Rt forecast is dated 2020-07-07 even though up-to-date timeseries are dated 2020-07-06
- This is because we noticed a bug in the submitted Rt forecast, so re-saved the forecast (bug: not all quantiles of US national were included in submission file)
- As of 2020-07-11, the "latest" Rt forecast is the 2020-07-07 version.
- In both the "latest" and "2020-07-07", the forecast_date variable is set to "2020-07-06" (as this is was when the forecast was actually made)