DataCollectorBase

class torchrl.collectors.collectors.DataCollectorBase(*args, **kwds)

Base class for data collectors.

update_policy_weights_(policy_weights: Optional[TensorDictBase] = None) → None

Updates the policy weights if the policy of the data collector and the trained policy live on different devices.

Parameters:

policy_weights (TensorDictBase, optional) – if provided, a TensorDict containing the weights of the policy to be used for the udpdate.