covid19-ensembles

In this repo we implement several ensemble methods for quantile forecasts and evaluate them on COVID-19 forecasts from the US Forecast Hub.

Overview:

• data_loading.R
  Used to load and preprocess the forecast files as well as the ensemble forecasts and truth data.

• data_exploration.R
  Here we explore forecast availability depending on target_end_date and number of locations. This is used to determine which models will be included in the ensembles.

• ensemble_methods.R
  Here we implement the different combination methods.

• ensemble_functions.R
  Utility functions to train the ensembles.

• ensemble_examples.R
  This contains some examples on how to use the methods provided in ensemble_methods.R (deprecated)

• compute_ensembles.R
  Here we train the ensembles and export their forecasts.

• scoring.R
  Contains some metrics (WIS score, absolute error), as well as utility functions to score forecasts.

• compute_scores.R
  Here we evaluate the individual and ensemble models and export their scores.

• evaluation_plots.R
  A collection of functions to visualize the results.