Trainer

class Melodie.trainer.Trainer(config: Config, scenario_cls: Type[Scenario] | None, model_cls: Type[Model] | None, data_loader_cls: Type[DataLoader] | None, processors: int = 1)

Bases: BaseModellingManager

Trainer trains the agents to update their behavioral parameters for higher payoff.

Parameters:

  • config – Config instance for current project.

  • scenario_cls – Scenario class for current project.

  • model_cls – Model class in current project.

  • data_loader_cls – DataLoader class in current project.

  • processors – Each path in current iteration will be computed parallelly, this parameter stands for processor cores used in parallel computation.

add_agent_training_property(agent_list_name: str, training_attributes: List[str], agent_ids: Callable[[Scenario], List[int]])

Add a container into the trainer.

Parameters:

  • agent_list_name – The name of agent container.

  • training_attributes – The properties used in training.

  • agent_ids – The agent with id contained in agent_ids will be trained.

Returns:

None

setup()

Setup method, be sure to inherit this method in custom trainer class.

get_trainer_scenario_cls()

Get the class of trainer scenario.

Returns:

Trainer parameters

collect_data()

Set the agent and environment properties to be collected.

Returns:

run()

The main method for Trainer.

Returns:

run_once_new(scenario: Scenario, trainer_params: GATrainerParams)

Run for one training path

Parameters:

  • scenario – The scenario to run

  • trainer_params – calibration parameters.

Returns:

None

utility(agent: Agent) → float

The utility is to be maximized. be sure to inherit inside the custom trainer class.

Parameters:

agent – Agent object.

Returns:

target_function(agent: Agent) → float

The target function to be minimized.

Parameters:

agent –

Returns:

add_agent_property(agent_list_name: str, prop: str)

Add a property of agent to be recorded.

Parameters:

  • agent_list_name – Name of agent list

  • prop – Property name of agent

add_environment_property(prop: str)

Add a property of environment to be recorded.

Parameters:

prop – Property name of environment.

generate_scenarios()

Generate Scenarios for trainer

Returns:

A list of scenario objects.

generate_trainer_params_list(trainer_scenario_cls: Type[GATrainerParams]) → List[GATrainerParams]

Generate Trainer Parameters.

Returns:

A list of trainer parameters.