This package implements several methods for epidemiological forecasting. It also implements recalibration methods for epidemiological forecasting, described in this paper.
See the README in the recalibration
folder.
The recalibration methods are implemented in epiforecast/R/calibrate.R
and scripts demonstrating their use on the FluSight Ensemble forecasters are located in calibration-experiments
.
To run the recalibration experiments, perform the following commands in the calibration-experiments
directory:
- Install the FluSight repo:
git clone https://github.com/FluSightNetwork/cdc-flusight-ensemble.git
- Create the component forecast file:
Rscript save_historical_forecasts.R
- Run desired
calibrate_*.R
scripts:calibrate_cv.R
recalibrates each of the FluSight Network component forecasters in an out-of-sample fashion.calibrate_ensemble.R
recalibrates the FluSight Network ensemble itself out-of-sample (see section 3.4).calibrate_sensitivity.R
recalibrates the FluSight Network component forecasters and experiments with different training window sizes (see sections 2.6, 3.1).calibrate_training.R
recalibrates the FluSight Network component forecasters and experiments with different numbers of training seasons (see section 3.3).calibrate_retro.R
recalibrates the FluSight Network component forecasters in a retrospective setting (see Supplementary Information section 3).