Adds description of param_noise parameter in deepq.learn method (#675)
This commit is contained in:
@@ -169,6 +169,8 @@ def learn(env,
|
||||
to 1.0. If set to None equals to total_timesteps.
|
||||
prioritized_replay_eps: float
|
||||
epsilon to add to the TD errors when updating priorities.
|
||||
param_noise: bool
|
||||
whether or not to use parameter space noise (https://arxiv.org/abs/1706.01905)
|
||||
callback: (locals, globals) -> None
|
||||
function called at every steps with state of the algorithm.
|
||||
If callback returns true training stops.
|
||||
|
Reference in New Issue
Block a user