mirror of
https://github.com/gryf/coach.git
synced 2025-12-17 11:10:20 +01:00
74 lines
3.6 KiB
Python
74 lines
3.6 KiB
Python
from rl_coach.agents.dfp_agent import DFPAgentParameters
|
|
from rl_coach.base_parameters import VisualizationParameters, EmbedderScheme, MiddlewareScheme, \
|
|
PresetValidationParameters
|
|
from rl_coach.core_types import EnvironmentSteps, RunPhase, EnvironmentEpisodes
|
|
from rl_coach.environments.doom_environment import DoomEnvironmentParameters
|
|
from rl_coach.environments.environment import SelectedPhaseOnlyDumpMethod, MaxDumpMethod
|
|
from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
|
|
from rl_coach.graph_managers.graph_manager import ScheduleParameters
|
|
from rl_coach.schedules import LinearSchedule
|
|
|
|
####################
|
|
# Graph Scheduling #
|
|
####################
|
|
|
|
schedule_params = ScheduleParameters()
|
|
schedule_params.improve_steps = EnvironmentSteps(6250000)
|
|
# original paper evaluates according to these. But, this preset converges significantly faster - can be evaluated
|
|
# much often.
|
|
# schedule_params.steps_between_evaluation_periods = EnvironmentSteps(62500)
|
|
# schedule_params.evaluation_steps = EnvironmentSteps(6250)
|
|
schedule_params.steps_between_evaluation_periods = EnvironmentEpisodes(5)
|
|
schedule_params.evaluation_steps = EnvironmentEpisodes(1)
|
|
|
|
# There is no heatup for DFP. heatup length is determined according to batch size. See below.
|
|
|
|
#########
|
|
# Agent #
|
|
#########
|
|
agent_params = DFPAgentParameters()
|
|
schedule_params.heatup_steps = EnvironmentSteps(agent_params.network_wrappers['main'].batch_size)
|
|
|
|
agent_params.network_wrappers['main'].learning_rate = 0.0001
|
|
agent_params.exploration.epsilon_schedule = LinearSchedule(0.5, 0, 10000)
|
|
agent_params.exploration.evaluation_epsilon = 0
|
|
agent_params.algorithm.goal_vector = [1] # health
|
|
|
|
# this works better than the default which is set to 8 (while running with 8 workers)
|
|
agent_params.algorithm.num_consecutive_playing_steps = EnvironmentSteps(1)
|
|
|
|
# scale observation and measurements to be -0.5 <-> 0.5
|
|
agent_params.network_wrappers['main'].input_embedders_parameters['measurements'].input_rescaling['vector'] = 100.
|
|
agent_params.network_wrappers['main'].input_embedders_parameters['measurements'].input_offset['vector'] = 0.5
|
|
agent_params.network_wrappers['main'].input_embedders_parameters['observation'].input_offset['vector'] = 0.5
|
|
|
|
# changing the network scheme to match Coach's default network, as it performs better on this preset
|
|
agent_params.network_wrappers['main'].input_embedders_parameters['observation'].scheme = EmbedderScheme.Medium
|
|
agent_params.network_wrappers['main'].input_embedders_parameters['measurements'].scheme = EmbedderScheme.Medium
|
|
agent_params.network_wrappers['main'].input_embedders_parameters['goal'].scheme = EmbedderScheme.Medium
|
|
agent_params.network_wrappers['main'].middleware_parameters.scheme = MiddlewareScheme.Medium
|
|
|
|
# scale the target measurements according to the paper (dividing by standard deviation)
|
|
agent_params.algorithm.scale_measurements_targets['GameVariable.HEALTH'] = 30.0
|
|
|
|
###############
|
|
# Environment #
|
|
###############
|
|
env_params = DoomEnvironmentParameters()
|
|
env_params.level = 'HEALTH_GATHERING'
|
|
vis_params = VisualizationParameters()
|
|
vis_params.video_dump_methods = [SelectedPhaseOnlyDumpMethod(RunPhase.TEST), MaxDumpMethod()]
|
|
vis_params.dump_mp4 = False
|
|
|
|
########
|
|
# Test #
|
|
########
|
|
preset_validation_params = PresetValidationParameters()
|
|
preset_validation_params.test = True
|
|
preset_validation_params.min_reward_threshold = 1600
|
|
preset_validation_params.max_episodes_to_achieve_reward = 70
|
|
|
|
graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,
|
|
schedule_params=schedule_params, vis_params=vis_params,
|
|
preset_validation_params=preset_validation_params)
|