Last updated . The n-day tabs show how the model prediction that was made n days ago compares to what actually happened over the past n days. The checkpoint tab shows how the model predictions that were made two weeks ago compare to what actually happened every day since, i.e., it is the 14-day score.
Benchmark models are selected models from our CS156 class this past spring, while the other models are new ones developed during our summer research. A simplistic baseline (all-zero prediction) is also included for reference. All errors are computed per county then averaged over all US counties. In addition, the models are combined in an ensemble and a novelty score for each model is computed.
|Rank||Model||Pinball loss||RMSE||Novelty score|