def test_observe_multiple_dimensions(self):
observer = action_metrics.ContinuousActionObserver()
observer.observe_first(env=_FAKE_ENV, timestep=_TIMESTEP)
observer.observe(env=_FAKE_ENV,
timestep=_TIMESTEP,
action=np.array([[1, 2], [3, 4]]))
np.testing.assert_equal(
{
'action[0, 0]_max': 1,
'action[0, 0]_min': 1,
'action[0, 0]_mean': 1,
'action[0, 0]_p50': 1,
'action[0, 1]_max': 2,
'action[0, 1]_min': 2,
'action[0, 1]_mean': 2,
'action[0, 1]_p50': 2,
'action[1, 0]_max': 3,
'action[1, 0]_min': 3,
'action[1, 0]_mean': 3,
'action[1, 0]_p50': 3,
'action[1, 1]_max': 4,
'action[1, 1]_min': 4,
'action[1, 1]_mean': 4,
'action[1, 1]_p50': 4,
},
observer.get_metrics(),
)
def test_observe_zero_dimensions(self):
observer = action_metrics.ContinuousActionObserver()
observer.observe_first(env=_FAKE_ENV, timestep=_TIMESTEP)
observer.observe(env=_FAKE_ENV, timestep=_TIMESTEP, action=np.array(1))
self.assertEqual(
{
'action[]_max': 1,
'action[]_min': 1,
'action[]_mean': 1,
'action[]_p50': 1,
},
observer.get_metrics(),
)
def test_observe_multiple_step(self):
observer = action_metrics.ContinuousActionObserver()
observer.observe_first(env=_FAKE_ENV, timestep=_TIMESTEP)
observer.observe(env=_FAKE_ENV,
timestep=_TIMESTEP,
action=np.array([1]))
observer.observe(env=_FAKE_ENV,
timestep=_TIMESTEP,
action=np.array([4]))
observer.observe(env=_FAKE_ENV,
timestep=_TIMESTEP,
action=np.array([5]))
self.assertEqual(
{
'action[0]_max': 5,
'action[0]_min': 1,
'action[0]_mean': 10 / 3,
'action[0]_p50': 4,
},
observer.get_metrics(),
)
def test_observe_first(self):
observer = action_metrics.ContinuousActionObserver()
observer.observe_first(env=_FAKE_ENV, timestep=_TIMESTEP)
self.assertEqual({}, observer.get_metrics())
def test_observe_nothing(self):
observer = action_metrics.ContinuousActionObserver()
self.assertEqual({}, observer.get_metrics())