gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2026-04-20 15:11:24 +02:00
Code Issues Projects Releases Wiki Activity

network_imporvements branch merge

Browse Source
This commit is contained in:
Shadi Endrawis
2018-10-02 13:41:46 +03:00
parent 72ea933384
commit 51726a5b80
110 changed files with 1639 additions and 1161 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rl_coach/presets/CartPole_PG.py
+6 -14
View File
@@ -1,9 +1,8 @@
from rl_coach.agents.policy_gradients_agent import PolicyGradientsAgentParameters
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps, RunPhase
from rl_coach.environments.environment import SelectedPhaseOnlyDumpMethod, MaxDumpMethod
from rl_coach.environments.gym_environment import MujocoInputFilter, Mujoco
from rl_coach.exploration_policies.categorical import CategoricalParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.gym_environment import GymVectorEnvironment
from rl_coach.filters.filter import InputFilter
from rl_coach.filters.reward.reward_rescale_filter import RewardRescaleFilter
from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
from rl_coach.graph_managers.graph_manager import ScheduleParameters
@@ -29,20 +28,13 @@ agent_params.algorithm.num_steps_between_gradient_updates = 20000
agent_params.network_wrappers['main'].optimizer_type = 'Adam'
agent_params.network_wrappers['main'].learning_rate = 0.0005
agent_params.input_filter = MujocoInputFilter()
agent_params.input_filter = InputFilter()
agent_params.input_filter.add_reward_filter('rescale', RewardRescaleFilter(1/200.))
agent_params.exploration = CategoricalParameters()
###############
# Environment #
###############
env_params = Mujoco()
env_params.level = 'CartPole-v0'
vis_params = VisualizationParameters()
vis_params.video_dump_methods = [SelectedPhaseOnlyDumpMethod(RunPhase.TEST), MaxDumpMethod()]
vis_params.dump_mp4 = False
env_params = GymVectorEnvironment(level='CartPole-v0')
########
# Test #
@@ -53,6 +45,6 @@ preset_validation_params.min_reward_threshold = 130
preset_validation_params.max_episodes_to_achieve_reward = 550
graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,
schedule_params=schedule_params, vis_params=vis_params,
schedule_params=schedule_params, vis_params=VisualizationParameters(),
preset_validation_params=preset_validation_params)