gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2026-03-02 06:35:47 +01:00
Code Issues Projects Releases Wiki Activity

removing some of the presets from the trace tests + more robust replay buffer loading

Browse Source
This commit is contained in:
itaicaspi-intel
2018-09-12 15:25:13 +03:00
parent 171fe97a3a
commit a16d724963
10 changed files with 59 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
rl_coach/presets/Atari_NEC.py
View File

@@ -39,7 +39,7 @@ vis_params.dump_mp4 = False
# Test #
########
preset_validation_params = PresetValidationParameters()
preset_validation_params.trace_test_levels = ['breakout', 'pong', 'space_invaders']
preset_validation_params.test_using_a_trace_test = False
graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,
schedule_params=schedule_params, vis_params=vis_params,

1
rl_coach/presets/CartPole_NEC.py
View File

@@ -50,6 +50,7 @@ preset_validation_params = PresetValidationParameters()
preset_validation_params.test = True
preset_validation_params.min_reward_threshold = 150
preset_validation_params.max_episodes_to_achieve_reward = 300
preset_validation_params.test_using_a_trace_test = False
graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,
schedule_params=schedule_params, vis_params=vis_params,

11
rl_coach/presets/Doom_Basic_BC.py
View File

@@ -1,5 +1,5 @@
from rl_coach.agents.bc_agent import BCAgentParameters
from rl_coach.base_parameters import VisualizationParameters
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.doom_environment import DoomEnvironmentParameters
from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
@@ -38,5 +38,12 @@ agent_params.memory.load_memory_from_file_path = 'datasets/doom_basic.p'
env_params = DoomEnvironmentParameters()
env_params.level = 'basic'
########
# Test #
########
preset_validation_params = PresetValidationParameters()
preset_validation_params.test_using_a_trace_test = False
graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,
schedule_params=schedule_params, vis_params=VisualizationParameters())
schedule_params=schedule_params, vis_params=VisualizationParameters(),
preset_validation_params=preset_validation_params)

1
rl_coach/presets/Doom_Health_DFP.py
View File

@@ -68,6 +68,7 @@ preset_validation_params.test = True
# reward threshold was set to 1000 since otherwise the test takes about an hour
preset_validation_params.min_reward_threshold = 1000
preset_validation_params.max_episodes_to_achieve_reward = 70
preset_validation_params.test_using_a_trace_test = False
graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,
schedule_params=schedule_params, vis_params=vis_params,

2
rl_coach/presets/Doom_Health_MMC.py
View File

@@ -45,7 +45,7 @@ vis_params.dump_mp4 = False
# Test #
########
preset_validation_params = PresetValidationParameters()
preset_validation_params.test_using_a_trace_test = False
# disabling this test for now, as it takes too long to converge
# preset_validation_params.test = True
# preset_validation_params.min_reward_threshold = 1000

11
rl_coach/presets/Doom_Health_Supreme_DFP.py
View File

@@ -1,5 +1,6 @@
from rl_coach.agents.dfp_agent import DFPAgentParameters
from rl_coach.base_parameters import VisualizationParameters, EmbedderScheme, MiddlewareScheme
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
@@ -59,6 +60,12 @@ 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_using_a_trace_test = False
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)

11
rl_coach/presets/MontezumaRevenge_BC.py
View File

@@ -1,5 +1,5 @@
from rl_coach.agents.bc_agent import BCAgentParameters
from rl_coach.base_parameters import VisualizationParameters
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
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 Atari
@@ -39,5 +39,12 @@ 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_using_a_trace_test = False
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=vis_params,
preset_validation_params=preset_validation_params)