Skip to content

Target Updates

lerax.target.AbstractTargetUpdate

Bases: eqx.Module

Rule for refreshing a target network from an online network.

__call__ abstractmethod 

__call__[T: eqx.Module](online: T, target: T) -> T

Return the updated target network.

lerax.target.SoftUpdate

Bases: AbstractTargetUpdate

Apply Polyak averaging.

target = tau * online + (1 - tau) * target.

Attributes:

Name Type Description
tau float

Interpolation coefficient in [0, 1]. Defaults to 0.005.

tau class-attribute instance-attribute 

tau: float = 0.005

__call__ 

__call__[T: eqx.Module](online: T, target: T) -> T

lerax.target.HardUpdate

Bases: AbstractTargetUpdate

Hard copy.

__call__ 

__call__[T: eqx.Module](online: T, target: T) -> T