Model Refitting using Entire Dataset

[isachng93]

One query: After selecting the best model based on test set accuracy, would the selected model be refitted using the entire dataset, inclusive of the test data?

--> In a similar way to what Modeltime packages does as described here: https://business-science.github.io/modeltime/reference/modeltime_refit.html

Thanks!

[winedarksea]

Yes, it uses the entire dataset for the final forecast. That is done in the .predict part. The .fit part only looks at the past in cross validation.

If you are looking at running AutoTS, say, weekly, I recommend you check out production_example.py in this repo. If you set evolve=False it will just run the same model (from template) each week, after initial training. With evolve=True there it will use the previous week as a base then try new additional models as well.

[winedarksea]

You can also use the model_forecast function with the best_model from the AutoTS class, it looks like the closest match to the model refit you showed there