HardUpdate

class torchrl.objectives.HardUpdate(loss_module: Union[DQNLoss, DDPGLoss, SACLoss, TD3Loss], *, value_network_update_interval: float = 1000)

A hard-update class for target network update in Double DQN/DDPG (by contrast with soft updates).

This was proposed in the original Double DQN paper: “Deep Reinforcement Learning with Double Q-learning”, https://arxiv.org/abs/1509.06461.

Parameters:

loss_module (DQNLoss or DDPGLoss) – loss module where the target network should be updated.

Keyword Arguments:

value_network_update_interval (scalar) – how often the target network should be updated. default: 1000