class EpsGreedyPolicyTest(unittest.TestCase):
def setUp(self):
self._eps = 0.1
self._acting = EpsGreedyPolicy(self._eps)
def test_seed(self):
seed = 123
eps = 0.7
lowest_q = 1
highest_q = 10
n_actions = 10
n_qs = 1000
acting1 = EpsGreedyPolicy(eps)
acting1.seed(seed)
acting2 = EpsGreedyPolicy(eps)
acting2.seed(seed)
qs = [
np.random.uniform(lowest_q, highest_q, size=(1, n_actions))
for _ in range(n_qs)
]
for q in qs:
a1 = acting1.act(q)
a2 = acting2.act(q)
self.assertEqual(a1, a2)
def test_act(self):
n_total = 10000
lowest_q = 1
highest_q = 10
n_actions = 100
n_max_q = 0
n_random = 0
for _ in range(n_total):
q = np.random.uniform(lowest_q, highest_q, size=(1, n_actions))
arg_max = np.argmax(q)
action = self._acting.act(q)
if arg_max == action:
n_max_q += 1
else:
n_random += 1
actual = n_random / n_total
max_deviation = 0.1
actual_deviation = abs((self._eps - actual) / self._eps)
self.assertLess(actual_deviation, max_deviation)
def test_seed(self):
seed = 123
eps = 0.7
lowest_q = 1
highest_q = 10
n_actions = 10
n_qs = 1000
acting1 = EpsGreedyPolicy(eps)
acting1.seed(seed)
acting2 = EpsGreedyPolicy(eps)
acting2.seed(seed)
qs = [
np.random.uniform(lowest_q, highest_q, size=(1, n_actions))
for _ in range(n_qs)
]
for q in qs:
a1 = acting1.act(q)
a2 = acting2.act(q)
self.assertEqual(a1, a2)