d3rlpy.augmentation.vector.MultipleAmplitudeScaling

class d3rlpy.augmentation.vector.MultipleAmplitudeScaling(minimum=0.8, maximum=1.2)

Multiple Amplitude Scaling augmentation.

\[x' = x + z\]

where \(z \sim \text{Unif}(minimum, maximum)\) and \(z\) is a vector with different amplitude scale on each.

References

• Laskin et al., Reinforcement Learning with Augmented Data.

Parameters

• minimum (float) – minimum amplitude scale.

• maximum (float) – maximum amplitude scale.

Methods

get_params(deep=False)

Returns augmentation parameters.

Parameters

deep (bool) – flag to copy parameters.

Returns

augmentation parameters.

Return type

Dict[str, Any]

get_type()

Returns augmentation type.

Returns

augmentation type.

Return type

str

transform(x)

Returns augmented observation.

Parameters

x (torch.Tensor) – observation.

Returns

augmented observation.

Return type

torch.Tensor

Attributes

TYPE: ClassVar[str] = 'multiple_amplitude_scaling'