`chronax.loss`

Loss functions for DeepAR model (JAX/FLAX implementation).

`nll_gaussian`

chronax.loss.nll_gaussian

Gaussian negative log-likelihood.

Parameter	Type	Default	Description
`y_true`	`jnp.ndarray`	-	Target values, shape (...,)
`mu`	`jnp.ndarray`	-	Mean predictions, shape (...,)
`sigma`	`jnp.ndarray`	-	Standard deviation predictions, shape (...,)

Returns: jnp.ndarray (NLL per element, shape (...,))

chronax.loss.nll_gaussian_masked

Masked Gaussian NLL (for handling variable-length horizons).

Parameter	Type	Default	Description
`y_true`	`jnp.ndarray`	-	Target values, shape [B, H, 1]
`mu`	`jnp.ndarray`	-	Mean predictions, shape [B, H]
`sigma`	`jnp.ndarray`	-	Std dev predictions, shape [B, H]
`mask`	`jnp.ndarray`	-	Binary mask, shape [B, H, 1], where 1 means include in loss

Returns: jnp.ndarray (Scalar loss (mean over unmasked elements))

chronax.loss.quantile_loss

Quantile loss for computing quantiles via quantile regression.

Parameter	Type	Default	Description
`y_true`	`jnp.ndarray`	-	Target values, shape (...,)
`y_pred`	`jnp.ndarray`	-	Predicted quantile values, shape (...,)
`quantile`	`float`	-	Quantile level in (0, 1)

Returns: jnp.ndarray (Quantile loss per element)