def testGetFilters(self):
ev = CommonPolicyEvaluator(
env_creator=lambda _: gym.make("CartPole-v0"),
policy_graph=MockPolicyGraph,
sample_async=True,
observation_filter="ConcurrentMeanStdFilter")
self.sample_and_flush(ev)
filters = ev.get_filters(flush_after=False)
time.sleep(2)
filters2 = ev.get_filters(flush_after=False)
obs_f = filters["default"]
obs_f2 = filters2["default"]
self.assertGreaterEqual(obs_f2.rs.n, obs_f.rs.n)
self.assertGreaterEqual(obs_f2.buffer.n, obs_f.buffer.n)
def testSyncFilter(self):
ev = CommonPolicyEvaluator(
env_creator=lambda _: gym.make("CartPole-v0"),
policy_graph=MockPolicyGraph,
sample_async=True,
observation_filter="ConcurrentMeanStdFilter")
obs_f = self.sample_and_flush(ev)
# Current State
filters = ev.get_filters(flush_after=False)
obs_f = filters["default"]
self.assertLessEqual(obs_f.buffer.n, 20)
new_obsf = obs_f.copy()
new_obsf.rs._n = 100
ev.sync_filters({"default": new_obsf})
filters = ev.get_filters(flush_after=False)
obs_f = filters["default"]
self.assertGreaterEqual(obs_f.rs.n, 100)
self.assertLessEqual(obs_f.buffer.n, 20)
def testFilterSync(self):
ev = CommonPolicyEvaluator(
env_creator=lambda _: gym.make("CartPole-v0"),
policy_graph=MockPolicyGraph,
sample_async=True,
observation_filter="ConcurrentMeanStdFilter")
time.sleep(2)
ev.sample()
filters = ev.get_filters(flush_after=True)
obs_f = filters["default"]
self.assertNotEqual(obs_f.rs.n, 0)
self.assertNotEqual(obs_f.buffer.n, 0)