The fully preprocessed dataset (M3Quart, Year 1980, MACRO, 52 Observations) is saved in the "M3C_processed.csv". This data is RNN prediction ready.
After doing predictions using RNNs, the inverse transformation is needed to convert normalized (scaled) values back to the original scale so they are interpretable in the context of the original data.
Information about the scaled data for the inverse transformation is saved in the "M3C_scaler_params.csv" file.
Our dataset contains the following columns: Series, N, NF, Category, Starting Year, Starting Quarter, and N observations (1,..., N).
- Series is just an identifier for the series itself (eg. N 1230)
- N is the number of observations for that series
- NF is the forecasting horizon for that series (how many observations we need to forecast). It is always 8 in the case of the M3Quart dataset we are using
- Category will always be constant, as we're only focusing on MACROs
- Starting Year is the starting year of the series
- Starting Quarter is the starting quarter of the series