agent_params.network_wrappers['main'].input_embedders_parameters['observation'] = \
InputEmbedderParameters(scheme=[Dense(200)])
agent_params.network_wrappers['main'].middleware_parameters = LSTMMiddlewareParameters(scheme=MiddlewareScheme.Empty,
number_of_lstm_cells=128)
agent_params.input_filter = InputFilter()
agent_params.input_filter.add_reward_filter('rescale', RewardRescaleFilter(1/20.))
agent_params.input_filter.add_observation_filter('observation', 'normalize', ObservationNormalizationFilter())
###############
# Environment #
###############
env_params = GymVectorEnvironment(level=SingleLevelSelection(mujoco_v2))
########
# Test #
########
preset_validation_params = PresetValidationParameters()
preset_validation_params.test = False
preset_validation_params.min_reward_threshold = 400
preset_validation_params.max_episodes_to_achieve_reward = 1000
preset_validation_params.num_workers = 8
preset_validation_params.reward_test_level = 'inverted_pendulum'
preset_validation_params.trace_test_levels = ['inverted_pendulum', 'hopper']
graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,
schedule_params=schedule_params, vis_params=VisualizationParameters(),
preset_validation_params=preset_validation_params)
agent_params.network_wrappers['critic'].input_embedders_parameters[
'observation'].scheme = EmbedderScheme.Empty
agent_params.network_wrappers['critic'].input_embedders_parameters[
'action'].scheme = EmbedderScheme.Empty
agent_params.network_wrappers['critic'].middleware_parameters.scheme = [
Dense(400), Dense(300)
]
###############
# Environment #
###############
env_params = GymVectorEnvironment(level=SingleLevelSelection(mujoco_v2))
########
# Test #
########
preset_validation_params = PresetValidationParameters()
preset_validation_params.test = True
preset_validation_params.min_reward_threshold = 500
preset_validation_params.max_episodes_to_achieve_reward = 1100
preset_validation_params.reward_test_level = 'hopper'
preset_validation_params.trace_test_levels = ['inverted_pendulum', 'hopper']
graph_manager = BasicRLGraphManager(
agent_params=agent_params,
env_params=env_params,
schedule_params=schedule_params,
vis_params=VisualizationParameters(),
preset_validation_params=preset_validation_params)