def test_sampling(self):
exp_q = NoReplacementQueue()
# obs r a e
exp_q.add((np.zeros(10), [1], [1], [0]))
exp_q.add((np.zeros(10), [0], [-1], [1])) # 1st episode end
exp_q.add((np.zeros(10), [1], [2], [0]))
exp_q.add((np.zeros(10), [1], [3], [0]))
exp_q.add((np.zeros(10), [1], [4], [0]))
exp_seqs = exp_q.sample(self.is_episode_end)
self.assertEqual(len(exp_q), 1)
self.assertEqual(len(exp_seqs), 2)
self.assertEqual(len(exp_seqs[0]), 2)
self.assertEqual(exp_seqs[0][0][2], [1])
self.assertEqual(exp_seqs[0][1][2], [-1])
self.assertEqual(len(exp_seqs[1]), 3)
self.assertEqual(exp_seqs[1][0][2], [2])
self.assertEqual(exp_seqs[1][1][2], [3])
self.assertEqual(exp_seqs[1][2][2], [4])
# obs r a e
exp_q.add((np.zeros(10), [0], [-2], [1]))
exp_seqs = exp_q.sample(self.is_episode_end)
self.assertEqual(len(exp_q), 0)
self.assertEqual(len(exp_seqs), 1)
self.assertEqual(len(exp_seqs[0]), 2)
self.assertEqual(exp_seqs[0][0][2], [4])
self.assertEqual(exp_seqs[0][1][2], [-2])
self.assertEqual(len(exp_q), 0)
class OnlineHelper(AgentHelper):
"""
Online helper. It calls `learn()` every `sample_interval`
steps.
While waiting for learning return, the calling `Agent` is blocked.
"""
def __init__(self, name, communicator, sample_interval=5):
super(OnlineHelper, self).__init__(name, communicator, sample_interval)
# NoReplacementQueue used to store past experience.
self.exp_queue = NoReplacementQueue()
@staticmethod
def exp_replay():
return False
def add_experience(self, e):
self.exp_queue.add(e)
def sample_experiences(self):
return self.exp_queue.sample()