gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2026-03-29 08:03:32 +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

21
rl_coach/presets/Atari_NStepQ.py
View File

@@ -1,8 +1,8 @@
from rl_coach.agents.n_step_q_agent import NStepQAgentParameters
from rl_coach.architectures.tensorflow_components.architecture import Conv2d, Dense
from rl_coach.architectures.tensorflow_components.layers import Conv2d, Dense
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps, RunPhase
from rl_coach.environments.environment import SingleLevelSelection, SelectedPhaseOnlyDumpMethod, MaxDumpMethod
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.environment import SingleLevelSelection
from rl_coach.environments.gym_environment import Atari, atari_deterministic_v4
from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
from rl_coach.graph_managers.graph_manager import ScheduleParameters
@@ -22,19 +22,14 @@ schedule_params.heatup_steps = EnvironmentSteps(0)
agent_params = NStepQAgentParameters()
agent_params.network_wrappers['main'].learning_rate = 0.0001
agent_params.network_wrappers['main'].input_embedders_parameters['observation'].scheme = [Conv2d([16, 8, 4]),
Conv2d([32, 4, 2])]
agent_params.network_wrappers['main'].middleware_parameters.scheme = [Dense([256])]
agent_params.network_wrappers['main'].input_embedders_parameters['observation'].scheme = [Conv2d(16, 8, 4),
Conv2d(32, 4, 2)]
agent_params.network_wrappers['main'].middleware_parameters.scheme = [Dense(256)]
###############
# Environment #
###############
env_params = Atari()
env_params.level = SingleLevelSelection(atari_deterministic_v4)
vis_params = VisualizationParameters()
vis_params.video_dump_methods = [SelectedPhaseOnlyDumpMethod(RunPhase.TEST), MaxDumpMethod()]
vis_params.dump_mp4 = False
env_params = Atari(level=SingleLevelSelection(atari_deterministic_v4))
########
# Test #
@@ -43,5 +38,5 @@ preset_validation_params = PresetValidationParameters()
preset_validation_params.trace_test_levels = ['breakout', 'pong', 'space_invaders']
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)