def test_epsilon(self):
actions = [10, 100, 1000]
pi = EpsilonGreedy(actions, QFunction(), 0.5)
self.assertEqual(0.5, pi.epsilon)
pi.epsilon = 0.6
self.assertEqual(0.6, pi.epsilon)
with self.assertRaises(AttributeError):
pi.epsilon = -1
with self.assertRaises(AttributeError):
pi.epsilon = 2
with self.assertRaises(AttributeError):
del pi.epsilon
def test_sample(self):
actions = [10, 100, 1000]
pi = EpsilonGreedy(actions, TestQFunction([10, 100, 1000]), 0)
self.assertEqual(1000, pi.sample_action(0))
self.assertTrue(
np.linalg.norm(np.array([0., 0., 1.]) - pi(0)) < 0.0000001)
pi.epsilon = 0.3
self.assertTrue(
np.linalg.norm(np.array([0.1, 0.1, 0.8]) - pi(0)) < 0.0000001)
pi.epsilon = 0.6
np.random.seed(0)
self.assertEqual(100, pi.sample_action(0))
pi = EpsilonGreedy(actions, TestQFunction([10, 1000, 100]), 0)
self.assertEqual(100, pi.sample_action(0))
self.assertTrue(
np.linalg.norm(np.array([0., 1., 0.]) - pi(0)) < 0.0000001)
pi.epsilon = 0.3
self.assertTrue(
np.linalg.norm(np.array([0.1, 0.8, 0.1]) - pi(0)) < 0.0000001)
pi.epsilon = 0.6
np.random.seed(0)
self.assertEqual(100, pi.sample_action(0))
pi = EpsilonGreedy(actions, TestQFunction([1000, 100, 10]), 0)
self.assertEqual(10, pi.sample_action(0))
self.assertTrue(
np.linalg.norm(np.array([1., 0., 0.]) - pi(0)) < 0.0000001)
pi.epsilon = 0.3
self.assertTrue(
np.linalg.norm(np.array([0.8, 0.1, 0.1]) - pi(0)) < 0.0000001)
pi.epsilon = 0.6
np.random.seed(0)
self.assertEqual(100, pi.sample_action(0))