def test_run_episode(self):
lal = linear_approx_learning.LinearApproxLearning()
lal._env.generate_starting_state = MagicMock(
return_value=State(Card(COLOR_BLACK, 1), 1))
lal.generate_action = MagicMock(return_value=ACTION_HIT)
lal._env.step = MagicMock(side_effect=[
(State(Card(COLOR_BLACK, 1), 1), 0),
(State(Card(COLOR_BLACK, 1), 1, is_terminal=True), 1),
])
lal.run_episode()
self.assertLess(
0.0, lal.evaluate_model(State(Card(COLOR_BLACK, 1), 1),
ACTION_HIT))
lal = linear_approx_learning.LinearApproxLearning()
lal._env.generate_starting_state = MagicMock(
return_value=State(Card(COLOR_BLACK, 1), 1))
lal.generate_action = MagicMock(return_value=ACTION_HIT)
lal._env.step = MagicMock(side_effect=[
(State(Card(COLOR_BLACK, 1), 1), 0),
(State(Card(COLOR_BLACK, 1), 1, is_terminal=True), -1),
])
lal.run_episode()
self.assertGreater(
0.0, lal.evaluate_model(State(Card(COLOR_BLACK, 1), 1),
ACTION_HIT))
def test_evaluate_model(self):
lal = linear_approx_learning.LinearApproxLearning()
lal._weights = np.array([.5] + [0] * 35)
self.assertEqual(
.5, lal.evaluate_model(State(Card(COLOR_BLACK, 1), 1), ACTION_HIT))
self.assertEqual(
0.0, lal.evaluate_model(State(Card(COLOR_BLACK, 11), 1),
ACTION_HIT))
def test_generate_action(self):
lal = linear_approx_learning.LinearApproxLearning()
lal.get_explore_threshold = MagicMock(return_value=0.0)
with patch.object(lal,
"evaluate_model",
new=lambda state, action: 1.0
if action == ACTION_HIT else 0.5):
self.assertEqual(ACTION_HIT, lal.generate_action(SAMPLE_STATE))
with patch.object(lal,
"evaluate_model",
new=lambda state, action: .5
if action == ACTION_HIT else 1.0):
self.assertEqual(ACTION_STICK, lal.generate_action(SAMPLE_STATE))
def test_get_explore_threshold(self):
lal = linear_approx_learning.LinearApproxLearning()
self.assertLessEqual(0.05, lal.get_explore_threshold(SAMPLE_STATE))
def test_init(self):
self.assertTrue(linear_approx_learning.LinearApproxLearning())