gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-17 19:20:19 +01:00
Code Issues Projects Releases Wiki Activity

Simulating the act on the trainer. (#65)

Browse Source 
* Remove the use of daemon threads for Redis subscribe.
* Emulate act and observe on trainer side to update internal vars.
This commit is contained in:
Ajay Deshpande
2018-11-15 08:38:58 -08:00
committed by Balaji Subramaniam
parent fe6857eabd
commit fde73ced13
13 changed files with 221 additions and 55 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
rl_coach/level_manager.py
View File

@@ -17,7 +17,7 @@ import copy
from typing import Union, Dict
from rl_coach.agents.composite_agent import CompositeAgent
from rl_coach.core_types import EnvResponse, ActionInfo, RunPhase, ActionType, EnvironmentSteps
from rl_coach.core_types import EnvResponse, ActionInfo, RunPhase, ActionType, EnvironmentSteps, Transition
from rl_coach.environments.environment import Environment
from rl_coach.environments.environment_interface import EnvironmentInterface
from rl_coach.spaces import ActionSpace, SpacesDefinition
@@ -264,5 +264,31 @@ class LevelManager(EnvironmentInterface):
def should_train(self) -> bool:
return any([agent._should_train_helper() for agent in self.agents.values()])
# TODO-remove - this is a temporary flow, used by the trainer worker, duplicated from observe() - need to create
# an external trainer flow reusing the existing flow and methods [e.g. observe(), step(), act()]
def emulate_step_on_trainer(self, transition: Transition) -> None:
"""
This emulates a step using the transition obtained from the rollout worker on the training worker
in case of distributed training.
Run a single step of following the behavioral scheme set for this environment.
:param action: the action to apply to the agents held in this level, before beginning following
the scheme.
:return: None
"""
if self.reset_required:
self.reset_internal_state()
acting_agent = list(self.agents.values())[0]
# for i in range(self.steps_limit.num_steps):
# let the agent observe the result and decide if it wants to terminate the episode
done = acting_agent.emulate_observe_on_trainer(transition)
acting_agent.emulate_act_on_trainer(transition)
if done:
self.handle_episode_ended()
self.reset_required = True
def should_stop(self) -> bool:
return all([agent.get_success_rate() >= self.environment.get_target_success_rate() for agent in self.agents.values()])