def testFromEpisodeArray(self):
observation = np.random.rand(4, 5)
action = ()
policy_info = ()
reward = np.random.rand(4)
traj = trajectory.from_episode(
observation, action, policy_info, reward, discount=None)
self.assertFalse(tf.contrib.framework.is_tensor(traj.step_type))
first = ts.StepType.FIRST
mid = ts.StepType.MID
last = ts.StepType.LAST
self.assertAllEqual(
traj.step_type, [first, mid, mid, mid])
self.assertAllEqual(
traj.next_step_type, [mid, mid, mid, last])
self.assertAllEqual(traj.observation, observation)
self.assertAllEqual(traj.reward, reward)
self.assertAllEqual(traj.discount, [1.0, 1.0, 1.0, 1.0])
def testFromEpisodeTensor(self):
observation = tf.random_uniform((4, 5))
action = ()
policy_info = ()
reward = tf.random_uniform((4,))
traj = trajectory.from_episode(
observation, action, policy_info, reward, discount=None)
self.assertTrue(tf.contrib.framework.is_tensor(traj.step_type))
traj_val, obs_val, reward_val = self.evaluate((traj, observation, reward))
first = ts.StepType.FIRST
mid = ts.StepType.MID
last = ts.StepType.LAST
self.assertAllEqual(
traj_val.step_type, [first, mid, mid, mid])
self.assertAllEqual(
traj_val.next_step_type, [mid, mid, mid, last])
self.assertAllEqual(traj_val.observation, obs_val)
self.assertAllEqual(traj_val.reward, reward_val)
self.assertAllEqual(traj_val.discount, [1.0, 1.0, 1.0, 1.0])