Structure of the repository

Our repository is structured as follows:

app_forecasts_de

Codes and re-formatted data underlying the interactive visualization.

code

• auto_download: Python codes to automate pulling some truth and forecast data to our repository (in progress)
• validation: Python codes for automatic validation of data format. We also provide R codes for the same purpose, but remark that these may not always be in perfect sync with the authoritative Python checks.

data-processed

Forecast files in a standard format, either processed from csv files in data_raw or directly submitted by teams via pull requests. Subdirectories within the data-processed/ directory need to have the format

team-model

where

• team is the teamname and
• model is the name of your model.

Both team and model should be less than 15 characters and not include hyphens.

Within each subdirectory, there should be a metadata file, a license file (optional), and a set of forecasts.

data_raw

The data-raw/ folder contains forecasts which have been made publicly available by research groups in their respective formats. R codes to process them to our standard format are stored in the respective subdirectories.

data-temp

This folder contains primarily forecasts for which processing into the standard format has been unsuccessful, but may be attempted again at a later time.

data-truth

Truth data from various sources, with ECDC currently being the encouraged truth data source within our project. Also contains R scripts to update truth data.

template

Templates of submission and metadata files along with some auxiliary data sets (e.g. on FIPS codes).

travis

Codes for automatic checks of submission using Travis CI.