Tweaking fastai tabular for time series data

You can make your model considerably more powerful by careful feature engineering for timeseries data.

Sources: - https://www.kaggle.com/learn/time-series - https://www.kaggle.com/competitions/playground-series-s5e1

Main ideas: - Add calendar periodic features - Need to combine test and train data together for this to work properly - Add autoregressive features - Augment with external data: - Currency exchange rate - Census (there are predictive data for this)