tsts.losses

Functions

class tsts.losses.build_losses[source]

Build a list of loss functions.

Parameters

cfg (CN) – Global configuration

Returns

List of loss functions

Return type

List[Loss]

Classes

class tsts.losses.DILATE[source]

DILATE implementation.

Example

LOSSES:
  NAMES: ["DILATE"]
  ARGS: [{"alpha": 0.5, "gamma": 0.001}]
Parameters

  • alpha (float, optional) – Balancing parameter of shape and temporal losses, by default 0.5

  • gamma (float, optional) – Smoothing parameter of softmin, by default 0.001

forward(Z: torch.Tensor, y: torch.Tensor, y_mask: torch.Tensor) → torch.Tensor[source]

Return loss value.

Parameters

  • Z (Tensor) – Prediction

  • y (Tensor) – Target time series

  • y_mask (Tensor) – Target time series mask

Returns

Loss value

Return type

Tensor

class tsts.losses.MAPE[source]

MAPE implementation.

Example

LOSSES:
  NAMES: ["MAPE"]
forward(Z: torch.Tensor, y: torch.Tensor, y_mask: torch.Tensor) → torch.Tensor[source]

Return loss value.

Parameters

  • Z (Tensor) – Prediction

  • y (Tensor) – Target time series

  • y_mask (Tensor) – Target time series mask

Returns

Loss value

Return type

Tensor

class tsts.losses.MSE[source]

MSE implementation.

Example

LOSSES:
  NAMES: ["MSE"]
forward(Z: torch.Tensor, y: torch.Tensor, y_mask: torch.Tensor) → torch.Tensor[source]

Return loss value.

Parameters

  • Z (Tensor) – Prediction

  • y (Tensor) – Target time series

  • y_mask (Tensor) – Target time series mask

Returns

Loss value

Return type

Tensor