def testAgent(RL, test, frames=2000):
test_env = tetrisML.TetrisGame("Testing " + test[0],
test[1],
test[2],
test[3],
log=True)
total_steps = 0
observation = test_env.reset()
while True:
actions_value = RL.get_reading(observation, kp=1)[0]
action = np.argmax(actions_value)
if action == 40 and not test_env.canUseHold: # Prevent it from using hold when not available
action = np.argsort(actions_value)[-2]
observation_, reward = test_env.nextFrame(action)
if total_steps > frames: # stop game
break
observation = observation_
total_steps += 1
return np.average(test_env.scores), np.average(
test_env.gamelengths), np.average(test_env.scoreChanges), np.average(
test_env.heuristicChanges)
def testRandom(test, frames=2000):
test_env = tetrisML.TetrisGame("Testing " + test[0],
test[1],
test[2],
test[3],
log=True)
total_steps = 0
observation = test_env.reset()
while True:
action = np.random.randint(0, 41)
observation_, reward = test_env.nextFrame(action)
if total_steps > frames: # stop game
break
observation = observation_
total_steps += 1
return np.average(test_env.scores), np.average(
test_env.gamelengths), np.average(test_env.scoreChanges), np.average(
test_env.heuristicChanges)
total_steps += 1
return np.average(test_env.scores), np.average(
test_env.gamelengths), np.average(test_env.scoreChanges), np.average(
test_env.heuristicChanges)
tests = [["Controle", True, True, 6], ["Onzichtbaar veld", False, True, 6],
["Onzichtbare hold", True, False, 6],
["Een next piece zichtbaar", True, True, 1],
["Geen next piece zichtbaar", True, True, 0]]
folder = path.dirname(__file__)
for test in tests:
env = tetrisML.TetrisGame("Training " + test[0], test[1], test[2], test[3])
MEMORY_SIZE = 100000
ACTION_SPACE = env.num_actions
FEATURES = env.num_features
FEATURESHAPE = env.featureShape
STATESHAPE = env.stateShape
sess = tf.Session()
with tf.variable_scope('Double_DQN'):
DQN = DeepQNetwork(n_actions=ACTION_SPACE,
n_features=FEATURES,
memory_size=MEMORY_SIZE,
e_greedy_increment=0.0000045,
e_greedy=0.9,