reference/generated/torchrl.modules.SymLogValueTransform

Run in Google Colab

Colab

Download Notebook

Notebook

View on GitHub

GitHub

SymLogValueTransform

class torchrl.modules.SymLogValueTransform(*args: Any, **kwargs: Any)

Symmetric-log value transform used by DreamerV3.

This transform applies torchrl.modules.functional.symlog() in the forward direction and torchrl.modules.functional.symexp() in the inverse direction.

Examples

>>> import torch
>>> from torchrl.modules import SymLogValueTransform
>>> transform = SymLogValueTransform()
>>> value = torch.tensor([-100.0, 0.0, 100.0])
>>> transformed = transform(value)
>>> transformed
tensor([-4.6151,  0.0000,  4.6151])
>>> transform.inverse(transformed)
tensor([-100.0000,    0.0000,  100.0000])

See also

torchrl.modules.functional.symlog() and torchrl.modules.functional.symexp() for the functional form, and SignedHyperbolicValueTransform for an alternative nonlinear transform.

Note

See Mastering Diverse Domains through World Models (Hafner et al., 2023).

forward(value: Tensor) → Tensor

Apply torchrl.modules.functional.symlog() to value.

inverse(value: Tensor) → Tensor

Apply torchrl.modules.functional.symexp() to value.