gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2026-03-28 22:53:44 +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

21
rl_coach/presets/Mujoco_NAF.py
View File

@@ -1,9 +1,9 @@
from rl_coach.agents.naf_agent import NAFAgentParameters
from rl_coach.architectures.tensorflow_components.architecture import Dense
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps, RunPhase, GradientClippingMethod
from rl_coach.environments.environment import MaxDumpMethod, SelectedPhaseOnlyDumpMethod, SingleLevelSelection
from rl_coach.environments.gym_environment import Mujoco, mujoco_v2
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps, GradientClippingMethod
from rl_coach.environments.environment import SingleLevelSelection
from rl_coach.environments.gym_environment import GymVectorEnvironment, mujoco_v2
from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
from rl_coach.graph_managers.graph_manager import ScheduleParameters
@@ -20,20 +20,15 @@ schedule_params.heatup_steps = EnvironmentSteps(1000)
# Agent #
#########
agent_params = NAFAgentParameters()
agent_params.network_wrappers['main'].input_embedders_parameters['observation'].scheme = [Dense([200])]
agent_params.network_wrappers['main'].middleware_parameters.scheme = [Dense([200])]
agent_params.network_wrappers['main'].input_embedders_parameters['observation'].scheme = [Dense(200)]
agent_params.network_wrappers['main'].middleware_parameters.scheme = [Dense(200)]
agent_params.network_wrappers['main'].clip_gradients = 1000
agent_params.network_wrappers['main'].gradients_clipping_method = GradientClippingMethod.ClipByValue
###############
# Environment #
###############
env_params = Mujoco()
env_params.level = SingleLevelSelection(mujoco_v2)
vis_params = VisualizationParameters()
vis_params.video_dump_methods = [SelectedPhaseOnlyDumpMethod(RunPhase.TEST), MaxDumpMethod()]
vis_params.dump_mp4 = False
env_params = GymVectorEnvironment(level=SingleLevelSelection(mujoco_v2))
# this preset is currently broken - no test
@@ -51,5 +46,5 @@ 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=vis_params,
schedule_params=schedule_params, vis_params=VisualizationParameters(),
preset_validation_params=preset_validation_params)