Skip to content

Transitioning from the US COVID19 Forecast Hub

Johannes Bracher edited this page Mar 1, 2021 · 6 revisions

This overview shall help you get started quickly if you are already familiar with the procedures for the US COVID-19 Forecast Hub. In case you notice an aspect that is missing please let us know by opening an issue.

What stays the same:

  • The basic folder structure with a model-specific subfolder in data-processed (naming convention team-model).
  • The forecast file format with point forecasts and 23 forecast quantiles.
  • Case and death forecasts will be collected.
  • The use of epi weeks running from Sunday to Saturday.
  • The metadata format.
  • The submission via GitHub pull requests (you just need to fork the covid19-forecast-hub-europe repository).
  • All evaluation criteria which will be applied to your forecasts (weighted interval score, absolute error, interval coverage rates).
  • The authoritative truth data source is JHU.
  • The weekly submission deadline on Monday at 6pm ET (midnight CET).

What changes:

  • We cover national-level forecasts for 32 European countries. The countries to be included and their mapping to location codes can be found here. Note that location codes are based on the ISO 3166-1 alpha-2 standard rather than FIPS codes.
  • Only incidence forecasts will be covered (i.e. k wk ahead inc death and k wk ahead inc case). Cumulative targets will not be covered.
  • No hospitalization/ICU forecasts will be accepted.
  • An (optional) additional column called scenario_id can be included. If included, this needs to be set to "forecast" in all rows for short-term forecasts. Other scenarios (e.g., on vaccination strategies) may be added in the future.