noisy_game_no_longer_valid = False
#Create testing enviroment
settings = {
'Side Engines': True,
'Clouds': True,
'Vectorized Nozzle': True,
'Starting Y-Pos Constant': 1,
'Initial Force': 'random'
} # (6000, -10000)}
env = RocketLander(settings)
env.refresh(render=True)
env.render(mode="human")
env.reset()
print("-- Observations", env.observation_space)
print("-- actionspace", env.action_space)
#initialize training matrix with random states and actions
#dataX = np.random.random(( 5,num_env_variables+num_env_actions ))
dataX = tf.placeholder("float", [None, num_env_variables + num_env_actions])
#Only one output for the total score / reward
#dataY = np.random.random((5,1))
dataY = tf.placeholder("float", [None, 1])
#initialize training matrix with random states and actions
#apdataX = np.random.random(( 5,num_env_variables ))
apdataX = tf.placeholder("float", [None, num_env_variables])
print("env.action_space", len(env.action_space))
print("action_bounds", action_bounds)
PG = PolicyGradient(
n_x = env.observation_space.shape[0],
n_y = len(env.action_space),
learning_rate=0.001,
reward_decay=0.99,
load_path=load_path,
save_path=save_path,
epsilon_max=0.95,
epsilon_greedy_increment=0.01,
initial_epsilon = 0.8
)
observation = env.reset()
left_or_right_barge_movement = np.random.randint(0, 2)
epsilon = 0.05
for episode in range(EPISODES):
while True:
# 1. Choose an action based on observation
action = PG.choose_action(observation)
# 2. Take action in the environment
observation_, reward, done, info = env.step(action)
# 3. Store transition for training
# if reward > -0.20: