gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2026-03-23 11:03:32 +01:00
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

17
rl_coach/presets/InvertedPendulum_PG.py
View File

@@ -1,8 +1,8 @@
from rl_coach.agents.policy_gradients_agent import PolicyGradientsAgentParameters
from rl_coach.base_parameters import VisualizationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps, RunPhase
from rl_coach.environments.environment import MaxDumpMethod, SelectedPhaseOnlyDumpMethod
from rl_coach.environments.gym_environment import Mujoco, MujocoInputFilter
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.observation.observation_normalization_filter import ObservationNormalizationFilter
from rl_coach.filters.reward.reward_rescale_filter import RewardRescaleFilter
from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
@@ -25,7 +25,7 @@ agent_params.algorithm.apply_gradients_every_x_episodes = 5
agent_params.algorithm.num_steps_between_gradient_updates = 20000
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(1/20.))
agent_params.input_filter.add_observation_filter('observation', 'normalize', ObservationNormalizationFilter())
@@ -33,14 +33,9 @@ agent_params.input_filter.add_observation_filter('observation', 'normalize', Obs
###############
# Environment #
###############
env_params = Mujoco()
env_params.level = "InvertedPendulum-v2"
vis_params = VisualizationParameters()
vis_params.video_dump_methods = [SelectedPhaseOnlyDumpMethod(RunPhase.TEST), MaxDumpMethod()]
vis_params.dump_mp4 = False
env_params = GymVectorEnvironment(level="InvertedPendulum-v2")
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())