RobustScaler

chronax.data.RobustScaler

Median + MAD scaler with std-based fallback when MAD == 0.

Stateless: stats returns (shift, scale) so the same values can be reused to inverse-transform predictions.

scaler = RobustScaler()
shift, scale = scaler.stats(insample, axis=1)   # [B,L] -> [B,1]
z    = scaler.transform(insample, shift, scale)
pred = scaler.inverse(z_hat, shift, scale)

stats(self, x: jnp.ndarray, axis: int = 1) -> Tuple[jnp.ndarray, jnp.ndarray]

Returns: Tuple[jnp.ndarray, jnp.ndarray]

transform(self, x: jnp.ndarray, shift: jnp.ndarray, scale: jnp.ndarray) -> jnp.ndarray

Returns: jnp.ndarray

inverse(self, z: jnp.ndarray, shift: jnp.ndarray, scale: jnp.ndarray) -> jnp.ndarray

Returns: jnp.ndarray

build_windows

chronax.data.build_windows

Return [n_windows, input_size + h] rolling windows (stride = 1).

Returns: jnp.ndarray Raises: ValueError: When T < input_size + h.

split_train_val_windows

chronax.data.split_train_val_windows

Chronological split: earliest windows train, latest windows validate.

When val_fraction <= 0 all windows go to the training set.

Returns: Tuple[jnp.ndarray, jnp.ndarray]

pad_sequence

chronax.data.pad_sequence

Left-pad / left-truncate a 1-D series to target_len.

Returns (padded, mask) where mask is 1 on real observations.

Returns: Tuple[jnp.ndarray, jnp.ndarray]

align_covariates

chronax.data.align_covariates

Align per-series exogenous arrays to model input expectations.

hist_exog → left-padded to [input_size, X]. futr_exog → left-padded to [input_size + h, F]. stat_exog → cast to float32, returned as-is.

Returns: Dict[str, Optional[jnp.ndarray]]

create_batch

chronax.data.create_batch

Build a single JAX batch from a list of complete time series.

Each series is split at the -h cut point. History is left-padded to input_size (real observations tracked by available_mask); horizon is padded to h (loss tracked by sample_mask).

Returns: Dict[str, Optional[jnp.ndarray]] (Dict with keys: insample_y [B, L, 1], outsample_y [B, h, 1], available_mask [B, L, 1], sample_mask [B, h, 1], hist_exog [B, L, X] or None, futr_exog [B, L+h, F] or None, stat_exog [B, S] or None)

batch_generator

chronax.data.batch_generator

Iterate over y_series yielding fully-prepared JAX batches.

Returns: Iterator[Dict[str, Optional[jnp.ndarray]]]