rlightning.policy.supervised_policy

class rlightning.policy.supervised_policy.SimpleSupervisedPolicy(config, role_type)

Bases: BasePolicy

A simple supervised policy with a CNN encoder for testing purposes.

construct_network(env_meta=None, *args, **kwargs)

Construct the neural network architecture.

Parameters:

• env_meta – Environment metadata for network configuration.

• *args – Variable positional arguments.

• **kwargs – Variable keyword arguments.

eval(obs)

forward(obs)

get_action(obs)

get_action_mean(obs)

get_action_value(obs)

get_trainable_parameters()

Return a dict of module state dicts.

get_value(obs)

init_train(train_config=None, env_meta=None)

Initialize the policy for training.

Sets up the network, finds trainable models, wraps with DDP if needed, and initializes the optimizer.

Parameters:

• train_config – Training configuration.

• env_meta – Environment metadata.

is_ready()

load_state_dict(state_dict, strict=True, assign=False)

Load trainable parameters from state_dict.

postprocess(env_ret=None, policy_resp=None)

rollout_step(env_ret, **kwargs)

setup_optimizer(optim_cfg)

Set up the optimizer for training.

Parameters:

optim_cfg – Optimizer configuration.

train(sl_buffer: DataBuffer)

Run a training step for the policy.

update_dataset(data)

Update internal dataset from sampled buffer data.