Compare the predictions of existing mortality forecasting models for Belgium, following what the CDC does for the US - https://www.cdc.gov/coronavirus/2019-ncov/covid-data/forecasting-us.html. For the US, the forecasts from around 15 models are collected every week by the Reich Lab
- Predictions can be visualised at https://reichlab.io/covid19-forecast-hub/
- Predictions are uploaded every Monday on their Github repository by the model developers https://github.com/reichlab/covid19-forecast-hub
For the moment, this repository provides:
- A notebook - Data-Collection.ipynb, that collects predictions from the YYG and IHME forecasting models from the 11/04 to the 23/05
- A Shiny interface to visualise the predictions, and assess/compare the model forecasting errors
TODO:
- Extend the data collection to more models
- Collect predictions for new hospitalisations
Truth data are located in the data-truth
folder. The following file is expected:
truth-incident-deaths
: Daily number of deaths. The format is- date : Date in YYYY-MM-DD format
- location: ID for the region, or 'All' if data for the whole country
- value: Number of deaths for the date in the region/country
The format follows the one suggested by ReichLab - https://github.com/reichlab/covid19-forecast-hub/blob/master/data-processed/README.md#Data-formatting
The file must be a comma-separated value (csv) file with the following columns (in any order):
- forecast_date
- target
- target_end_date
- location
- type
- quantile
- value
The data collection notebook currently collects data for the YYG and IHME models.
Reich forecasts only provided for US data. Belgium data is in https://github.com/youyanggu/covid19_projections/blob/master/projections/2020-05-21/global/Belgium_ALL.csv
Archived predictions are available from http://www.healthdata.org/covid/data-downloads.
- Methodology and code substantially designed thanks to Reich Lab - https://github.com/reichlab/covid19-forecast-hub