Github actions

[sbfnk]

https://github.com/epiforecasts/covid19-forecast-hub-europe/actions

• [X] Visualization/Validation/Evaluation This: 1) creates visualisation for a Shiny app and RKI viz 2) creates a static plot (which we don't need) 3) gets commit dates (not sure why) and performs data validation (currently deactivated) 4) evaluates forecasts (can be done once a week) To be reviewed / confirmed with people from the Germany/Poland forecast hub how much of this is really needed
• [X] ~~MZ This downloads subnational data for Poland - not needed~~
• [X] ~~PR checks This performs the validation checks on submissions - we need it~~
• [X] ~~epiforecasts This pulls (or will eventually pull) epiforecasts forecasts - not needed~~
• [X] ~~RKI-data This downloads subnational data for Poland - not needed~~
• [X] ~~UCLA This downloads UCLA forecasts - not needed~~
• [X] ~~LANL This downloads LANL forecasts - not needed~~
• [X] ~~Geneva This downloads Geneva forecasts - not needed~~
• [X] ~~ECDC-JHU-DIVI This downloads ECDC/JHU data - we will need some version of this~~

[jbracher]

If you want to use the RKI viz you do need the step called "creates visualisation for a Shiny app" above. The respective codes and output csvs are still in the folder of the Shiny app because this is what they were initially used for, but the viz set up by our colleagues at RKI uses the same csv files.

Specifically you need to generate the following:

• files in the format of truth_to_plot_ecdc.csv and truth_to_plot_jhu for truth data (or actually you'll need to get rid of one of them, but you need to discuss with Fabian from RKI how to do that exactly). In our repo prepare_truth_data.py does that.
• files in the format of forecasts_to_plot.csv for forecasts. You can probably get rid of some columns truth_data_source, shift_ECDC, shift_JHU, first_commit_date, but this will of course require adaptations from Fabian's side, otherwise the app will break. These files are currently generated by data_preparation.py and add_last_observed.py. The former does most of the work, the latter just adds the last observed values for each forecast, which is needed for plotting (I can't remember why there are different files for that). Especially data_preparation.py can be simplified considerably as some parts (columns truth_data_source, shift_ECDC, shift_JHU don't need to be generated as you only have one truth data source). The file data_preparation.R is no longer relevant.

Daniel Wolffram has written these Python scripts, he's also on the Slack channel in case you have questions.

[sbfnk]

Might also actually want the actions that pull data from elsewhere as discussed elsewhere (to populate the platform if/when not many models are available).

[jbracher]

Visualization/Validation/Evaluation is one large block because these things are just run jointly on a regular basis. Evaluation would only be necessary once per week. Commit dates are likely not needed anymore. For a while we thought we'd mark late or corrected forecasts in the app.