TargetNetMixin

class TargetNetMixin(target_update_freq: int = 0)

Bases: object

Mixin that adds a target network to the agent.

Duplicates the estimator networks that are used to estimate targets. These clones are updated at a target_update_freq frequency which allows to stabilize the training process and make targets less non-stationary.

References

“Playing Atari with Deep Reinforcement Learning”

https://arxiv.org/pdf/1312.5602.pdf

Methods Summary

sync_target(self)

Methods Documentation

sync_target(self)