SutraNets: Sub-series Autoregressive Networks for Long-Sequence, Probabilistic Forecasting: Supplementary Materials

These values are normalized, as noted above, using min-max values from the conditioning range; the prediction range can go below the min and above the max.