I have a doubt regarding the model and how this works during model evaluation.
Option 1 [fake forecasting]:
The model trains its weights on the training dataset and, during evaluation, it performs the prediction of the current sample by looking at the previous sample and back. In this case, we are just predicting 1 sample at a time, having all the true values of the previous ones.
Option 2 [prophet forecasting]:
In this case, the model trains its weights on the training dataset and, during evaluation, it just computes the future values without having any idea of the test dataset. The test dataset is just used to calculate the RMSE, thus we do not need to submit the model, because the forecasting is independent of the test dataset.
Which one is correct in this case?
Posted by: EugenioBertolini @ Dec. 21, 2021, 2:21 p.m.The second option is the correct one. However, in order to be evaluated you must use the submission standard given in the instructions, since we need to keep trace of your models.
Posted by: an2dl.competitions @ Dec. 23, 2021, 11:58 a.m.