gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2026-04-20 23:41:24 +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/ExplorationChain_UCB_Q_ensembles.py
+5 -9
View File
@@ -1,7 +1,7 @@
from rl_coach.agents.bootstrapped_dqn_agent import BootstrappedDQNAgentParameters
from rl_coach.base_parameters import VisualizationParameters
from rl_coach.core_types import EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.gym_environment import Mujoco
from rl_coach.environments.gym_environment import GymVectorEnvironment
from rl_coach.exploration_policies.ucb import UCBParameters
from rl_coach.filters.filter import NoInputFilter, NoOutputFilter
from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
@@ -30,8 +30,8 @@ agent_params.network_wrappers['main'].learning_rate = 0.00025
agent_params.memory.max_size = (MemoryGranularity.Transitions, 1000000)
agent_params.algorithm.discount = 0.99
agent_params.algorithm.num_consecutive_playing_steps = EnvironmentSteps(4)
agent_params.network_wrappers['main'].num_output_head_copies = num_output_head_copies
agent_params.network_wrappers['main'].rescale_gradient_from_head_by_factor = [1.0/num_output_head_copies]*num_output_head_copies
agent_params.network_wrappers['main'].heads_parameters[0].num_output_head_copies = num_output_head_copies
agent_params.network_wrappers['main'].heads_parameters[0].rescale_gradient_from_head_by_factor = 1.0/num_output_head_copies
agent_params.exploration = UCBParameters()
agent_params.exploration.bootstrapped_data_sharing_probability = 1.0
agent_params.exploration.architecture_num_q_heads = num_output_head_copies
@@ -43,12 +43,8 @@ agent_params.output_filter = NoOutputFilter()
###############
# Environment #
###############
env_params = Mujoco()
env_params.level = 'rl_coach.environments.toy_problems.exploration_chain:ExplorationChain'
env_params = GymVectorEnvironment(level='rl_coach.environments.toy_problems.exploration_chain:ExplorationChain')
env_params.additional_simulator_parameters = {'chain_length': N, 'max_steps': N+7}
vis_params = VisualizationParameters()
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())