unreal.LearningAgentsTrainerSettings

class unreal.LearningAgentsTrainerSettings

Bases: StructBase

The configurable settings for a ULearningAgentsTrainer.

C++ Source:

  • Plugin: LearningAgents

  • Module: LearningAgentsTraining

  • File: LearningAgentsTrainer.h

Editor Properties: (see get_editor_property/set_editor_property)

  • max_episode_step_num (int32): [Read-Write] Maximum number of steps recorded in an episode before it is added to the replay buffer. This can generally be left at the default value and does not have a large impact on training.

  • maximum_recorded_episodes_per_iteration (int32): [Read-Write] Maximum number of episodes to record before running a training iteration. An iteration of training will be run when either this or MaximumRecordedEpisodesPerIteration is reached. Typical values for this should be around 1000. Setting this too small means there is not enough data each iteration for the system to train. Setting it too large means training will be very slow.

  • maximum_recorded_steps_per_iteration (int32): [Read-Write] Maximum number of steps to record before running a training iteration. An iteration of training will be run when either this or MaximumRecordedEpisodesPerIteration is reached. Typical values for this should be around 10000. Setting this too small means there is not enough data each iteration for the system to train. Setting it too large means training will be very slow.

  • trainer_communication_timeout (float): [Read-Write] Time in seconds to wait for the training subprocess before timing out.