`tft_scaler`

Per-window scalers for TFT: identity (NF default) and robust (median/MAD).

Self-contained (no cross-model import). Pure functions exposing (stats, transform, inverse) so the same shift/scale is reused for inverse.

`Scaler`

tft_scaler.Scaler · implements Protocol

A runtime checkable protocol defining the interface for per-window scalers.

`stats(self, x, axis=1)`

Calculates the shift and scale parameters required for transformation.

Parameter	Type	Default	Description
`x`	`jnp.ndarray`	-	(undocumented)
`axis`	`int`	`1`	(undocumented)

Returns: tuple[jnp.ndarray, jnp.ndarray] (shift, scale).

`transform(self, x, shift, scale)`

Applies the scaling transformation.

Parameter	Type	Default	Description
`x`	`jnp.ndarray`	-	(undocumented)
`shift`	`jnp.ndarray`	-	(undocumented)
`scale`	`jnp.ndarray`	-	(undocumented)

Returns: jnp.ndarray (the transformed array).

`inverse(self, z, shift, scale)`

Applies the inverse scaling transformation.

Parameter	Type	Default	Description
`z`	`jnp.ndarray`	-	(undocumented)
`shift`	`jnp.ndarray`	-	(undocumented)
`scale`	`jnp.ndarray`	-	(undocumented)

Returns: jnp.ndarray (the inverse transformed array).

`IdentityScaler`

tft_scaler.IdentityScaler

No-op scaler (shift=0, scale=1). Matches neuralforecast's scaler_type='identity'.

`stats(self, x, axis=1)`

Calculates the shift (zeros) and scale (ones) parameters.

Parameter	Type	Default	Description
`x`	`jnp.ndarray`	-	(undocumented)
`axis`	`int`	`1`	(undocumented)

Returns: tuple[jnp.ndarray, jnp.ndarray] (shift, scale).

`transform(self, x, shift, scale)`

Returns the input array unchanged.

Parameter	Type	Default	Description
`x`	`jnp.ndarray`	-	(undocumented)
`shift`	`jnp.ndarray`	-	(undocumented)
`scale`	`jnp.ndarray`	-	(undocumented)

Returns: jnp.ndarray.

`inverse(self, z, shift, scale)`

Returns the input array unchanged.

Parameter	Type	Default	Description
`z`	`jnp.ndarray`	-	(undocumented)
`shift`	`jnp.ndarray`	-	(undocumented)
`scale`	`jnp.ndarray`	-	(undocumented)

Returns: jnp.ndarray.

`RobustScaler`

tft_scaler.RobustScaler

Median + MAD scaler with 0.6745*std fallback when MAD=0.

`stats(self, x, axis=1)`

Calculates the median (shift) and MAD (scale), using a scaled standard deviation fallback if MAD is zero.

Parameter	Type	Default	Description
`x`	`jnp.ndarray`	-	(undocumented)
`axis`	`int`	`1`	(undocumented)

Returns: tuple[jnp.ndarray, jnp.ndarray] (median, scale).

`transform(self, x, shift, scale)`

Applies robust scaling: (x - shift) / scale.

Parameter	Type	Default	Description
`x`	`jnp.ndarray`	-	(undocumented)
`shift`	`jnp.ndarray`	-	(undocumented)
`scale`	`jnp.ndarray`	-	(undocumented)

Returns: jnp.ndarray.

`inverse(self, z, shift, scale)`

Applies inverse robust scaling: z * scale + shift.

Parameter	Type	Default	Description
`z`	`jnp.ndarray`	-	(undocumented)
`shift`	`jnp.ndarray`	-	(undocumented)
`scale`	`jnp.ndarray`	-	(undocumented)

Returns: jnp.ndarray.

`resolve_scaler`

tft_scaler.resolve_scaler(scaler)

Resolve a scaler from a name ('identity'/'robust') or a Scaler instance.

Parameter	Type	Default	Description
`scaler`	`str \\| Scaler`	-	(undocumented)

Returns: Scaler.

Raises: * ValueError: If an unknown scaler name is provided.