metadata-COVIDhub-ensemble.txt

team_name: COVID-19 Forecast Hub
model_name: ensemble
model_abbr: COVIDhub-ensemble
model_contributors: Nutcha Wattanachit, Evan L. Ray, Nicholas Reich <nick@schoolph.umass.edu>
website_url: https://covid19forecasthub.org/
license: cc-by-4.0
team_model_designation: primary
ensemble_of_hub_models: true
methods: An ensemble, or model average, of submitted forecasts to the COVID-19 Forecast
  Hub.
team_funding: This model's development is funded by the US CDC grant number U01IP001122. The content is solely the responsibility of the authors and does not necessarily represent the official views of NIGMS or the National Institutes of Health.
repo_url: https://github.com/reichlab/covid19-forecast-hub
twitter_handles: reichlab, jaradniemi
data_inputs: cumulative and incident death forecasts from other models.
methods_long: 'April 13: At state level, only two models included in first week, LANL
  and IHME, due to overlaps in location and reported quantiles. At national level,
  IHME was combined with the three CU models. April 27: Different component models
  at state level and national level. Please see the model information file for more
  details. May 4: National ensemble combined 7 models, state ensembles combined between
  2 and 8 models. For details, see ensemble-metadata/state-weight-information.csv
  All ensembles were equally weighted among all eligible models. May 19: See ensemble-metadata/2020-05-19-model-eligibility.csv
  for models included/excluded at each location with reasons, and ensemble-metadata/2020-05-19-model-weights.csv
  for weights by location. May 26: See ensemble-metadata folder for models included/excluded
  in each location, and weights by location. July 28: switched to using a per-quantile
  median prediction across all eligible models. For cumulative deaths, we still implemented
  the automated checks that the 10th percentile of the 1-week ahead predictive distribution
  must be at least the most recent observed cumulative death, and the quantiles must be
  non-decreasing over time at each quantile level. We did not eliminate any models based on
  visual misalignment with the reference JHU data this week.'