Figure 7
From: Accurate long-range forecasting of COVID-19 mortality in the USA
Partitioning the dataset into train, validation and test sets. The forecasting task is to use the covariates at weeks t, t-1, t-2 to forecast the target variable at week t + 9. So the forecast horizon is nine (r = 9) and the history length is three (h = 3). Each row in the graph corresponds to an instance. The length of the dashed arrows are the forecast horizon and the length of the rectangles are the history length. The last instance is taken as the test dataset; that is, to forecast week October 3–10 using the covariates on weeks July 25 to August 8. The earlier eight red instances are not used and the yellow one before is used as the validation data instance. The green instances form the training dataset.
