def OnEpisodeFinishedCallback(self, env: Environment, brain: Brain,
episode_idx: int, num_of_episodes: int,
episode_reward: float, steps: int):
values = brain.GetValues(self._states)
for idx, v in enumerate(values):
for a in self._actions:
self._value_traces[a][idx].append(v[a])
def Decide(
self,
env: base.Environment,
brain: base.Brain,
state: base.State,
episode_idx: int,
num_of_episodes: int,
) -> base.Action:
return env.GetActionFromChoice(
numpy.random.choice(env.GetActionSpaceSize(),
p=brain.GetValues(state)[0]))
def Decide(
self,
env: base.Environment,
brain: base.Brain,
state: base.State,
episode_idx: int,
num_of_episodes: int,
) -> base.Action:
values = brain.GetValues(state)
choice = int(numpy.argmax(values))
logging.vlog(
20,
'making greedy decision for state %s using values: %s; choice: %d',
state, values, choice)
return env.GetActionFromChoice(choice)