Scaling

ns_sdn.typing.Scaling = Scaling

Alias of Literal['minmax', 'none'].

How the sample is scaled before training.

"minmax" maps the sample to \([-1, 1]\),

\[\tilde y_t = 2\,\frac{y_t - \min y}{\max y - \min y} - 1,\]

with constants the result keeps and inverts, so every reported path is on the original scale unless asked for in training units; the penalty weights and the Huber threshold are then interpretable across series. "none" trains on the raw series, which is the right choice only when the series is already of unit order.