unilab.algos.mlx.common.normalization.EmpiricalDiscountedVariationNormalization
-
class unilab.algos.mlx.common.normalization.EmpiricalDiscountedVariationNormalization[source]
Bases: object
Reward normalization with running std of discounted returns.
- Parameters:
-
Methods
-
__init__(eps=0.01, gamma=0.99, dtype=None)[source]
- Parameters:
-
-
__call__(rew)[source]
Normalize reward tensor of shape [N] or [N, 1].
- Parameters:
rew (array)
- Return type:
array