learner

class xuance.mindspore.learners.learner.Learner(config: Namespace, policy: mindspore.nn.Cell, callback)

Bases: ABC

estimate_total_iterations()

Estimated total number of training iterations

get_grad_reducer(optimizer: dict | list | mindspore.nn.Optimizer | None) → mindspore.nn.DistributedGradReducer | None

load_model(path, model=None)

save_model(model_path)

abstractmethod update(*args)

class xuance.mindspore.learners.learner.LearnerMAS(config: Namespace, model_keys: List[str], agent_keys: List[str], policy: mindspore.nn.Cell, callback)

Bases: ABC

build_training_data(sample: dict | None, use_parameter_sharing: bool | None = False, use_actions_mask: bool | None = False, use_global_state: bool | None = False)

Prepare the training data.

Parameters:

• sample (dict) – The raw sampled data.

• use_parameter_sharing (bool) – Whether to use parameter sharing for individual agent models.

• use_actions_mask (bool) – Whether to use actions mask for unavailable actions.

• use_global_state (bool) – Whether to use global state.

Returns:

The formatted sampled data.

Return type:

sample_Tensor (dict)

estimate_total_iterations()

Estimated total number of training iterations

get_joint_input(input_tensor, output_shape=None)

load_model(path, model=None)

save_model(model_path)

abstractmethod update(*args)

update_rnn(*args)