distance_loss¶
- class torchrl.objectives.distance_loss(v1: TensorLike, v2: TensorLike, loss_function: str, strict_shape: bool = True)[source]¶
Computes a distance loss between two tensors.
- Parameters:
v1 (Tensor | TensorDict) – a tensor or tensordict with a shape compatible with v2.
v2 (Tensor | TensorDict) – a tensor or tensordict with a shape compatible with v1.
loss_function (str) – One of “l2”, “l1” or “smooth_l1” representing which loss function is to be used.
strict_shape (bool) – if False, v1 and v2 are allowed to have a different shape. Default is
True.
- Returns:
A tensor or tensordict of the shape v1.view_as(v2) or v2.view_as(v1)
with values equal to the distance loss between the two.
