Question: CH04 - Is the test set part of the train set?

[dominicdill]

I am confused by the use of train and test sets for these time series models.

Here you setup your entire dataset and split it into test and train sets.

It seems like you would want to evaluate your model by seeing how well it performs on the test set using only the train set. But in the notebook the test set is used when evaluating the model.

To actually see how well the model predicts into the future, wouldn't we instead want to set the df parameter of the rolling_forcast function to train and then add lines to that function that extend the supplied df with the predictions?

Currently, because df_dff (which holds both train and test data) is used, you're not truly seeing how well your models predict into the future.

Am I misunderstanding? Is this what we want to be doing?

I changed the code to do as I think I'm expecting:

And I see this: