hive.runners.base module

class hive.runners.base.Runner(environment, agents, logger, experiment_manager, train_steps, eval_environment=None, test_frequency=-1, test_episodes=1, max_steps_per_episode=1000000000.0)

Bases: ABC, Registrable

Base Runner class used to implement a training loop.

Different types of training loops can be created by overriding the relevant functions.

Parameters

• environment (BaseEnv) – Environment used in the training loop.

• agents (list[Agent]) – List of agents that interact with the environment.

• logger (ScheduledLogger) – Logger object used to log metrics.

• experiment_manager (Experiment) – Experiment object that saves the state of the training.

• train_steps (int) – How many steps to train for. This is the number of times that agent.update is called. If this is -1, there is no limit for the number of training steps.

• test_frequency (int) – After how many training steps to run testing episodes. If this is -1, testing is not run.

• test_episodes (int) – How many episodes to run testing for.

register_config(config)

train_mode(training)

If training is true, sets all agents to training mode. If training is false, sets all agents to eval mode.

Parameters

training (bool) – Whether to be in training mode.

create_episode_metrics()

Create the metrics used during the loop.

update_step()

Update steps for various schedules. Run testing if appropriate.

run_episode(environment)

Run a single episode of the environment.

Parameters

environment (BaseEnv) – Environment in which the agent will take a step in.

run_training()

Run the training loop. Note, to ensure that the test phase is run during the individual runners must call update_step() in their run_episode() methods. See SingleAgentRunner and MultiAgentRunner for examples.

run_testing()

Run a testing phase.

resume()

Resume a saved experiment.

classmethod type_name()

Returns

“runner”