def test_shaping_reward_is_shaping(self):
base_reward = (1, 2)
shaping_reward = (3, 4)
reward = Reward(base_reward, shaping_reward)
self.assertTrue(reward.is_shaping())
def test_base_reward_is_not_shaping(self):
base_reward = (1, 2)
shaping_reward = ()
reward = Reward(base_reward, shaping_reward)
self.assertFalse(reward.is_shaping())
def test_shaping_reward_non_shaping_reward(self):
base_reward = (1, 2)
shaping_reward = (3, 4)
reward = Reward(base_reward, shaping_reward)
expected_non_shaping_r = sum(base_reward) / len(base_reward)
self.assertAlmostEqual(expected_non_shaping_r, reward.assessment_reward())
def _store_reward(self, reward: rewards.Reward, sim: Simulation):
sim[self.last_agent_reward] = reward.agent_reward()
sim[self.last_assessment_reward] = reward.assessment_reward()
def assess(self, state: State, prev_state: State, is_terminal: bool) -> Reward:
""" Calculates a Reward from the state transition. """
return Reward(self._base_rewards(state, prev_state, is_terminal),
self._potential_based_rewards(state, prev_state, is_terminal))
def assess(self, state: State, prev_state: State,
is_terminal: bool) -> Reward:
base_reward = (0, )
shaping_reward = ()
return Reward(base_reward, shaping_reward)
def test_init_error_on_empty_base_reward(self):
empty_base_reward = ()
shaping_reward = (1, 2)
with self.assertRaises(ValueError):
_ = Reward(empty_base_reward, shaping_reward)