def test_save_restore(self):
# Set up training data
td = training_data.training_data()
td.add(np.ones([1, 4, 4]), 0, 4, np.zeros([1, 4, 4]))
td.add(np.zeros([1, 4, 4]), 1, 2, np.ones([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 2, 16, np.zeros([1, 4, 4]))
td.add(np.zeros([1, 4, 4]), 3, 2, np.ones([1, 4, 4]))
temp_dir = tempfile.mkdtemp()
temp_filename = os.path.join(temp_dir, 'data.csv')
td.export_csv(temp_filename)
td2 = training_data.training_data()
td2.import_csv(temp_filename)
expected_x = np.array(
[[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]],
[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]]],
dtype=np.int)
expected_y_digit = np.array([[0], [1], [2], [3]], dtype=np.int)
expected_reward = np.array([[4], [2], [16], [2]], dtype=np.float)
expected_next_x = np.array(
[[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]],
[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]],
[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]]],
dtype=np.int)
assert np.array_equal(td2.get_x(), expected_x)
assert np.array_equal(td2.get_y_digit(), expected_y_digit)
assert np.allclose(td2.get_reward(), expected_reward)
assert np.array_equal(td2.get_next_x(), expected_next_x)
os.remove(temp_filename)
os.rmdir(temp_dir)
def test_smooth_rewards(self):
# Set up training data
td = training_data.training_data()
td.add(np.ones([1, 4, 4]), 0, 4, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 1, 2, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 2, 16, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 3, 2, np.zeros([1, 4, 4]))
# Test using default lambda value of 0.9
td2 = td.copy()
td2.smooth_rewards()
expected_reward = np.array([[20.218], [18.02], [17.8], [2.0]],
dtype=np.float)
self.assertTrue(np.allclose(td2.get_reward(), expected_reward))
# Test using lambda value of 0, should have no effect on rewards
td2 = td.copy()
td2.smooth_rewards(llambda=0.0)
expected_reward = np.array([[4], [2], [16], [2]], dtype=np.float)
self.assertTrue(np.allclose(td2.get_reward(), expected_reward))
# Test end of episode
td3 = training_data.training_data()
td3.add(np.ones([1, 4, 4]), 0, 4, np.zeros([1, 4, 4]), False)
td3.add(np.ones([1, 4, 4]), 1, 2, np.zeros([1, 4, 4]), True)
td3.add(np.ones([1, 4, 4]), 2, 16, np.zeros([1, 4, 4]), False)
td3.add(np.ones([1, 4, 4]), 3, 2, np.zeros([1, 4, 4]), True)
td3.smooth_rewards()
expected_reward = np.array([[5.8], [2.0], [17.8], [2.0]],
dtype=np.float)
self.assertTrue(np.allclose(td3.get_reward(), expected_reward))
def test_save_restore(self):
# Set up training data
td = training_data.training_data()
td.add(np.ones([1, 4, 4]), 0, 4)
td.add(np.zeros([1, 4, 4]), 1, 2)
td.add(np.ones([1, 4, 4]), 2, 16)
td.add(np.zeros([1, 4, 4]), 3, 2)
f = tempfile.NamedTemporaryFile()
td.export_csv(f.name)
td2 = training_data.training_data()
td2.import_csv(f.name)
expected_x = np.array(
[[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]],
[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]]],
dtype=np.int)
expected_y_digit = np.array([[0], [1], [2], [3]], dtype=np.int)
expected_reward = np.array([[4], [2], [16], [2]], dtype=np.float)
self.assertTrue(np.array_equal(td2.get_x(), expected_x))
self.assertTrue(np.array_equal(td2.get_y_digit(), expected_y_digit))
self.assertTrue(np.allclose(td2.get_reward(), expected_reward))
os.remove(f.name)
def test_normalize_rewards(self):
# Test calculating mean and standard deviation
td = training_data.training_data()
td.add(np.ones([1, 4, 4]), 1, 4, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 2, 4, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 3, 8, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 0, 16, np.zeros([1, 4, 4]))
td.normalize_rewards()
expected_reward = np.array([
[-0.8165],
[-0.8165],
[0.],
[1.633],
],
dtype=np.float)
self.assertTrue(np.allclose(td.get_reward(), expected_reward))
# Test specifying mean and standard deviation
td = training_data.training_data()
td.add(np.ones([1, 4, 4]), 1, 4, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 2, 4, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 3, 8, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 0, 16, np.zeros([1, 4, 4]))
td.normalize_rewards(mean=8, sd=1)
expected_reward = np.array([
[-4.],
[-4.],
[0.],
[8.],
],
dtype=np.float)
self.assertTrue(np.allclose(td.get_reward(), expected_reward))
def test_merge(self):
td = training_data.training_data()
td.add(np.ones([1, 4, 4]), 1, 16, np.zeros([1, 4, 4]))
td2 = training_data.training_data()
td2.add(np.zeros([1, 4, 4]), 2, 0, np.ones([1, 4, 4]))
td.merge(td2)
expected_x = np.array([
[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]]
], dtype=int)
expected_y_digit = np.array([
[1],
[2]
], dtype=int)
expected_reward = np.array([
[16],
[0]
], dtype=float)
expected_next_x = np.array([
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]],
[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]]
], dtype=int)
assert np.array_equal(td.get_x(), expected_x)
assert np.array_equal(td.get_y_digit(), expected_y_digit)
assert np.allclose(td.get_reward(), expected_reward)
assert np.array_equal(td.get_next_x(), expected_next_x)
def test_get_discounted_return(self):
# Set up training data
td = training_data.training_data()
td.add(np.ones([1, 4, 4]), 0, 4, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 1, 2, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 2, 16, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 3, 2, np.zeros([1, 4, 4]))
# Test using default gamma value of 0.9
td2 = td.copy()
discounted_return = td2.get_discounted_return()
expected_return = np.array([[20.218], [18.02], [17.8], [2.0]],
dtype=np.float)
assert np.allclose(discounted_return, expected_return)
# Test using gamma value of 0, should have no effect on rewards
td2 = td.copy()
discounted_return = td2.get_discounted_return(gamma=0.0)
expected_return = np.array([[4], [2], [16], [2]], dtype=np.float)
assert np.allclose(discounted_return, expected_return)
# Test end of episode
td3 = training_data.training_data()
td3.add(np.ones([1, 4, 4]), 0, 4, np.zeros([1, 4, 4]), False)
td3.add(np.ones([1, 4, 4]), 1, 2, np.zeros([1, 4, 4]), True)
td3.add(np.ones([1, 4, 4]), 2, 16, np.zeros([1, 4, 4]), False)
td3.add(np.ones([1, 4, 4]), 3, 2, np.zeros([1, 4, 4]), True)
discounted_return = td3.get_discounted_return()
expected_return = np.array([[5.8], [2.0], [17.8], [2.0]],
dtype=np.float)
assert np.allclose(discounted_return, expected_return)
def test_split(self):
td = training_data.training_data()
td.add(np.ones([1, 4, 4]), 1, 16)
td2 = training_data.training_data()
td2.add(np.zeros([1, 4, 4]), 2, 0)
td.merge(td2)
a, b = td.split()
self.assertTrue(np.array_equal(a.get_x(), np.ones([1, 4, 4])))
self.assertTrue(np.array_equal(a.get_y_digit(), [[1]]))
self.assertTrue(np.array_equal(b.get_x(), np.zeros([1, 4, 4])))
self.assertTrue(np.array_equal(b.get_y_digit(), [[2]]))
def test_split(self):
td = training_data.training_data()
td.add(np.ones([1, 4, 4]), 1, 16, np.zeros([1, 4, 4]))
td2 = training_data.training_data()
td2.add(np.zeros([1, 4, 4]), 2, 0, np.ones([1, 4, 4]))
td.merge(td2)
a, b = td.split()
assert np.array_equal(a.get_x(), np.ones([1, 4, 4]))
assert np.array_equal(a.get_y_digit(), [[1]])
assert np.array_equal(a.get_reward(), [[16]])
assert np.array_equal(a.get_next_x(), np.zeros([1, 4, 4]))
assert np.array_equal(b.get_x(), np.zeros([1, 4, 4]))
assert np.array_equal(b.get_y_digit(), [[2]])
assert np.array_equal(b.get_reward(), [[0]])
assert np.array_equal(b.get_next_x(), np.ones([1, 4, 4]))
def add_rewards_to_training_data(env, input_training_data):
new_training_data = training_data.training_data()
for n in range(input_training_data.size()):
(state, action) = input_training_data.get_n(n)
reward = get_reward_for_state_action(env, state, action)
new_training_data.add(state, action, reward)
return new_training_data
def test_hflip(self):
td = training_data.training_data()
board1 = np.array([[1, 1, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0],
[0, 0, 0, 0]])
board2 = np.array([[0, 0, 0, 0], [2, 4, 0, 0], [0, 0, 0, 0],
[0, 0, 0, 0]])
td.add(board1, 1, 2, board2)
td.add(board2, 2, 0, board1)
td.hflip()
expected_x = np.array(
[[[0, 0, 1, 1], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]],
[[0, 0, 0, 0], [0, 0, 4, 2], [0, 0, 0, 0], [0, 0, 0, 0]]],
dtype=np.int)
expected_y_digit = np.array([[3], [2]], dtype=np.int)
expected_reward = np.array([
[2],
[0],
], dtype=np.float)
expected_next_x = np.array(
[[[0, 0, 0, 0], [0, 0, 4, 2], [0, 0, 0, 0], [0, 0, 0, 0]],
[[0, 0, 1, 1], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]]],
dtype=np.int)
assert np.array_equal(td.get_x(), expected_x)
assert np.array_equal(td.get_y_digit(), expected_y_digit)
assert np.allclose(td.get_reward(), expected_reward)
assert np.allclose(td.get_next_x(), expected_next_x)
def test_normalize_boards(self):
# Test calculating mean and standard deviation
td = training_data.training_data()
td.add(np.full((1, 4, 4), 4), 1, 4, np.full((1, 4, 4), 8))
td.add(np.full((1, 4, 4), 8), 2, 4, np.full((1, 4, 4), 16))
td.add(np.full((1, 4, 4), 16), 3, 4, np.full((1, 4, 4), 32))
td.add(np.full((1, 4, 4), 32), 4, 4, np.full((1, 4, 4), 64))
td.normalize_boards()
mean = 15.
sd = 10.7238052947636
a = (4. - mean) / sd
b = (8. - mean) / sd
c = (16. - mean) / sd
d = (32. - mean) / sd
e = (64. - mean) / sd
expected_x = np.array(
[[[a, a, a, a], [a, a, a, a], [a, a, a, a], [a, a, a, a]],
[[b, b, b, b], [b, b, b, b], [b, b, b, b], [b, b, b, b]],
[[c, c, c, c], [c, c, c, c], [c, c, c, c], [c, c, c, c]],
[[d, d, d, d], [d, d, d, d], [d, d, d, d], [d, d, d, d]]],
dtype=np.float)
assert np.allclose(td.get_x(), expected_x)
expected_next_x = np.array(
[[[b, b, b, b], [b, b, b, b], [b, b, b, b], [b, b, b, b]],
[[c, c, c, c], [c, c, c, c], [c, c, c, c], [c, c, c, c]],
[[d, d, d, d], [d, d, d, d], [d, d, d, d], [d, d, d, d]],
[[e, e, e, e], [e, e, e, e], [e, e, e, e], [e, e, e, e]]],
dtype=np.float)
assert np.allclose(td.get_next_x(), expected_next_x)
def test_rotate(self):
td = training_data.training_data()
board1 = np.array([[1, 1, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0],
[0, 0, 0, 0]])
board2 = np.array([[0, 0, 0, 0], [2, 4, 0, 0], [0, 0, 0, 0],
[0, 0, 0, 0]])
td.add(board1, 1, 2, board2)
td.add(board2, 2, 0, board1)
td.rotate(3)
expected_x = np.array(
[[[0, 0, 0, 0], [0, 0, 0, 0], [1, 0, 0, 0], [1, 0, 0, 0]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 4, 0, 0], [0, 2, 0, 0]]],
dtype=np.int)
expected_y_digit = np.array([[0], [1]], dtype=np.int)
expected_reward = np.array([
[2],
[0],
], dtype=np.float)
expected_next_x = np.array(
[[[0, 0, 0, 0], [0, 0, 0, 0], [0, 4, 0, 0], [0, 2, 0, 0]],
[[0, 0, 0, 0], [0, 0, 0, 0], [1, 0, 0, 0], [1, 0, 0, 0]]],
dtype=np.int)
self.assertTrue(np.array_equal(td.get_x(), expected_x))
self.assertTrue(np.array_equal(td.get_y_digit(), expected_y_digit))
self.assertTrue(np.allclose(td.get_reward(), expected_reward))
self.assertTrue(np.array_equal(td.get_next_x(), expected_next_x))
def test_get_total_reward(self):
td = training_data.training_data()
td.add(np.ones([4, 4]), 0, 4, np.zeros([4, 4]))
td.add(np.zeros([4, 4]), 1, 8, np.ones([4, 4]))
td.add(np.zeros([4, 4]), 3, 16, np.ones([4, 4]))
td.add(np.zeros([4, 4]), 2, 32, np.ones([4, 4]))
assert td.get_total_reward() == 60
def test_get_n(self):
td = training_data.training_data()
td.add(np.ones([4, 4]), 1, 4, np.zeros([4, 4]))
td.add(np.zeros([4, 4]), 2, 8, np.ones([4, 4]))
(state, action, reward, next_state, done) = td.get_n(1)
assert np.array_equal(state, np.zeros([4, 4], dtype=np.int))
assert action == 2
assert reward == pytest.approx(8.)
assert np.array_equal(next_state, np.ones([4, 4], dtype=np.int))
def test_get_n(self):
# Test get_n with reward
td = training_data.training_data()
td.add(np.ones([4, 4]), 1, 4)
td.add(np.zeros([4, 4]), 2, 8)
(state, action, reward) = td.get_n(1)
self.assertTrue(np.array_equal(state, np.zeros([4, 4], dtype=np.int)))
self.assertEqual(action, 2)
self.assertAlmostEqual(reward, 8)
def test_augment(self):
td = training_data.training_data()
initial_board = np.array([[1, 1, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0]])
next_board = np.array([[0, 0, 0, 2],
[0, 2, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0]])
td.add(initial_board, 1, 4, next_board)
td.augment()
assert td.size() == 8
expected_x = np.array([
[[1, 1, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]],
[[0, 0, 1, 1], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]],
[[0, 0, 0, 1], [0, 0, 0, 1], [0, 0, 0, 0], [0, 0, 0, 0]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 1], [0, 0, 0, 1]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 1, 1]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [1, 1, 0, 0]],
[[0, 0, 0, 0], [0, 0, 0, 0], [1, 0, 0, 0], [1, 0, 0, 0]],
[[1, 0, 0, 0], [1, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]]
], dtype=int)
expected_y_digit = np.array([
[1],
[3],
[2],
[0],
[3],
[1],
[0],
[2]
], dtype=int)
expected_reward = np.array([
[4],
[4],
[4],
[4],
[4],
[4],
[4],
[4]
], dtype=float)
expected_next_x = np.array([
[[0, 0, 0, 2], [0, 2, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]], # Original
[[2, 0, 0, 0], [0, 0, 2, 0], [0, 0, 0, 0], [0, 0, 0, 0]], # Hflip'd
[[0, 0, 0, 0], [0, 0, 2, 0], [0, 0, 0, 0], [0, 0, 0, 2]], # Original, rotated 90 degrees
[[0, 0, 0, 2], [0, 0, 0, 0], [0, 0, 2, 0], [0, 0, 0, 0]], # Hflip, rotated 90 degrees
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 2, 0], [2, 0, 0, 0]], # Original, rotated 180 degrees
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 2, 0, 0], [0, 0, 0, 2]], # Hflip, rotated 180 degrees
[[2, 0, 0, 0], [0, 0, 0, 0], [0, 2, 0, 0], [0, 0, 0, 0]], # Original, rotate 270 degrees
[[0, 0, 0, 0], [0, 2, 0, 0], [0, 0, 0, 0], [2, 0, 0, 0]] # Hflip, rotated 270 degrees
], dtype=int)
assert np.array_equal(td.get_x(), expected_x)
assert np.array_equal(td.get_y_digit(), expected_y_digit)
assert np.allclose(td.get_reward(), expected_reward)
assert np.array_equal(td.get_next_x(), expected_next_x)
def test_get_x_stacked(self):
td = training_data.training_data()
td.add(np.full([4, 4], 2), 0, 4, np.zeros([4, 4]))
td.add(np.full([4, 4], 8), 1, 8, np.ones([4, 4]))
td.add(np.full([4, 4], 2048), 1, 8, np.ones([4, 4]))
expected_x_stacked = np.array(
[[[[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]],
[[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]],
[[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]],
[[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]],
[[[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]],
[[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]],
[[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]],
[[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]],
[[[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0]],
[[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0]],
[[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0]],
[[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0]]]],
dtype=np.int)
assert np.array_equal(td.get_x_stacked(), expected_x_stacked)
def test_get_n(self):
td = training_data.training_data()
td.add(np.ones([4, 4]), 1, 4, np.zeros([4, 4]))
td.add(np.zeros([4, 4]), 2, 8, np.ones([4, 4]))
(state, action, reward, next_state, done) = td.get_n(1)
self.assertTrue(np.array_equal(state, np.zeros([4, 4], dtype=np.int)))
self.assertEqual(action, 2)
self.assertAlmostEqual(reward, 8)
self.assertTrue(
np.array_equal(next_state, np.ones([4, 4], dtype=np.int)))
def test_get_y_one_hot(self):
td = training_data.training_data()
td.add(np.ones([4, 4]), 0, 4, np.zeros([4, 4]))
td.add(np.zeros([4, 4]), 1, 8, np.ones([4, 4]))
td.add(np.zeros([4, 4]), 3, 8, np.ones([4, 4]))
td.add(np.zeros([4, 4]), 2, 8, np.ones([4, 4]))
expected_y_one_hot = np.array(
[[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 0, 1], [0, 0, 1, 0]],
dtype=np.int)
assert np.array_equal(td.get_y_one_hot(), expected_y_one_hot)
def test_sample(self):
td = training_data.training_data()
td.add(np.zeros([1, 4, 4]), 0, 0, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 1, 1, np.ones([1, 4, 4]))
sample = td.sample([1])
assert sample.size() == 1
assert sample.get_y_digit() in [[[0]], [[1]]]
if sample.get_y_digit() == 0:
assert np.array_equal(sample.get_x(), np.zeros([1, 4, 4]))
if sample.get_y_digit() == 1:
assert np.array_equal(sample.get_x(), np.ones([1, 4, 4]))
def test_sample(self):
td = training_data.training_data()
td.add(np.zeros([1, 4, 4]), 0, 0, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 1, 1, np.ones([1, 4, 4]))
sample = td.sample([1])
self.assertEqual(sample.size(), 1)
self.assertIn(sample.get_y_digit(), [[[0]], [[1]]])
if sample.get_y_digit() == 0:
self.assertTrue(np.array_equal(sample.get_x(), np.zeros([1, 4,
4])))
if sample.get_y_digit() == 1:
self.assertTrue(np.array_equal(sample.get_x(), np.ones([1, 4, 4])))
def test_add(self):
td = training_data.training_data()
assert np.array_equal(td.get_x(), np.empty([0, 4, 4], dtype=int))
assert np.array_equal(td.get_y_digit(), np.empty([0, 1], dtype=int))
assert np.allclose(td.get_reward(), np.empty([0, 1], dtype=float))
assert np.array_equal(td.get_next_x(), np.empty([0, 4, 4], dtype=int))
assert np.array_equal(td.get_done(), np.empty([0, 1], dtype=bool))
td.add(np.ones([1, 4, 4]), 1, 4, np.zeros([1, 4, 4]), True)
assert np.array_equal(td.get_x(), np.ones([1, 4, 4], dtype=int))
assert np.array_equal(td.get_y_digit(), np.array([[1]], dtype=int))
assert np.allclose(td.get_reward(), np.array([[4]], dtype=float))
assert np.array_equal(td.get_next_x(), np.zeros([1, 4, 4], dtype=int))
assert np.array_equal(td.get_done(), np.array([[1]], dtype=bool))
def test_log2_rewards(self):
# Set up training data
td = training_data.training_data()
td.add(np.ones([1, 4, 4]), 0, 0)
td.add(np.ones([1, 4, 4]), 1, 2)
td.add(np.ones([1, 4, 4]), 2, 4)
td.add(np.ones([1, 4, 4]), 3, 16)
td.add(np.ones([1, 4, 4]), 0, 75)
td.add(np.ones([1, 4, 4]), 1, 2048)
td.log2_rewards()
expected_reward = np.array([[0], [1], [2], [4], [6.2288], [11]],
dtype=np.float)
self.assertTrue(np.allclose(td.get_reward(), expected_reward))
expected_action = np.array([[0], [1], [2], [3], [0], [1]],
dtype=np.int)
self.assertTrue(np.allclose(td.get_y_digit(), expected_action))
def test_add(self):
# Test add with reward
td = training_data.training_data()
self.assertTrue(
np.array_equal(td.get_x(), np.empty([0, 4, 4], dtype=np.int)))
self.assertTrue(
np.array_equal(td.get_y_digit(), np.empty([0, 1], dtype=np.int)))
self.assertTrue(
np.allclose(td.get_reward(), np.empty([0, 1], dtype=np.float)))
td.add(np.ones([1, 4, 4]), 1, 4)
self.assertTrue(
np.array_equal(td.get_x(), np.ones([1, 4, 4], dtype=np.int)))
self.assertTrue(
np.array_equal(td.get_y_digit(), np.array([[1]], dtype=np.int)))
self.assertTrue(
np.allclose(td.get_reward(), np.array([[4]], dtype=np.float)))
def test_log2_rewards(self):
# Set up training data
td = training_data.training_data()
td.add(np.ones([1, 4, 4]), 0, 0, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 1, 2, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 2, 4, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 3, 16, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 0, 75, np.zeros([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 1, 2048, np.zeros([1, 4, 4]))
td.log2_rewards()
expected_reward = np.array([
[0], [1], [2], [4], [6.2288], [11]
], dtype=float)
assert np.allclose(td.get_reward(), expected_reward)
expected_action = np.array([
[0], [1], [2], [3], [0], [1]
], dtype=int)
assert np.allclose(td.get_y_digit(), expected_action)
def test_shuffle(self):
td = training_data.training_data()
n = 5
for i in range(n):
# Use "is odd" for done
td.add(np.full((1, 4, 4), i), i, i, np.full((1, 4, 4), i), (i % 2) == 1)
td.shuffle()
for i in range(n):
# Find where this has been shuffled too
index_of_val = np.where(td.get_y_digit() == i)[0].item(0)
# Check that all parts of this equal i
arrays = td.get_n(index_of_val)
for a in arrays:
if a.dtype is np.dtype(bool):
assert((a == ((i % 2) == 1)).all())
else:
assert((a == i).all())
def test_make_boards_unique(self):
td = training_data.training_data()
td.add(np.ones([1, 4, 4]), 0, 4, np.zeros([1, 4, 4]))
td.add(np.zeros([1, 4, 4]), 1, 2, np.ones([1, 4, 4]))
td.add(np.ones([1, 4, 4]), 2, 16, np.zeros([1, 4, 4]))
td.add(np.zeros([1, 4, 4]), 3, 2, np.ones([1, 4, 4]))
td.make_boards_unique()
expected_x = np.array(
[[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]]],
dtype=np.int)
expected_y_digit = np.array([[0], [1]], dtype=np.int)
expected_reward = np.array([[4], [2]], dtype=np.float)
expected_next_x = np.array(
[[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]],
[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]]],
dtype=np.int)
assert np.array_equal(td.get_x(), expected_x)
assert np.array_equal(td.get_y_digit(), expected_y_digit)
assert np.allclose(td.get_reward(), expected_reward)
assert np.array_equal(td.get_next_x(), expected_next_x)
def gather_training_data(env, seed=None):
"""Gather training data from letting the user play the game"""
# Data is a list of input and outputs
data = training_data.training_data()
# Initialise seed for environment
if seed:
env.seed(seed)
else:
env.seed()
observation = env.reset()
print("User cursor keys to play, q to quit")
try:
while True:
# Loop around performing moves
action = None
env.render()
# Ask user for action
# Read and discard the keyup event
print("Read action {}".format(action))
# Add this data to the data collection
new_observation, reward, done, info = env.step(action)
if np.array_equal(observation, new_observation):
print("Suppressing recording of illegal move")
else:
data.add(observation, action, reward)
observation = new_observation
print()
if done:
# Draw final board
env.render()
print("End of game")
break
except Exiting:
print("Exiting...")
return data
def test_augment(self):
td = training_data.training_data()
initial_board = np.array([[1, 1, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0],
[0, 0, 0, 0]])
td.add(initial_board, 1, 4)
td.augment()
self.assertEqual(td.size(), 8)
expected_x = np.array(
[[[1, 1, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]],
[[0, 0, 1, 1], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]],
[[0, 0, 0, 1], [0, 0, 0, 1], [0, 0, 0, 0], [0, 0, 0, 0]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 1], [0, 0, 0, 1]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 1, 1]],
[[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [1, 1, 0, 0]],
[[0, 0, 0, 0], [0, 0, 0, 0], [1, 0, 0, 0], [1, 0, 0, 0]],
[[1, 0, 0, 0], [1, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]]],
dtype=np.int)
expected_y_digit = np.array([[1], [3], [2], [0], [3], [1], [0], [2]],
dtype=np.int)
expected_reward = np.array([[4], [4], [4], [4], [4], [4], [4], [4]],
dtype=np.float)
self.assertTrue(np.array_equal(td.get_x(), expected_x))
self.assertTrue(np.array_equal(td.get_y_digit(), expected_y_digit))
self.assertTrue(np.allclose(td.get_reward(), expected_reward))
import numpy as np
import face_detect as face_detect
import training_data as training_data
label = []
def predict(test_img):
img = cv2.imread(test_img).copy()
print "\n\n\n"
print "Face Prediction Running -\-"
face, rect, length = face_detect.face_detect(test_img)
print len(face), "faces detected."
for i in range(0, len(face)):
labeltemp, confidence = face_recognizer.predict(face[i])
label.append(labeltemp)
return img, label
faces, labels = training_data.training_data("training-data")
face_recognizer = cv2.face.LBPHFaceRecognizer_create()
face_recognizer.train(faces, np.array(labels))
# Read the test image.
test_img = "test-data/test.jpg"
predicted_img, label = predict(test_img)
cv2.destroyAllWindows()
cv2.waitKey(1)
cv2.destroyAllWindows()
print "Recognized faces = ", label
