def main():
gym_env = gym.make('custom_gym:Xplane-v0')
lr = 0.001
gam = 0.01
n_games = 1
# nn_input = obs()
agent = Agent(learning_rate=lr,
gamma=gam,
epsilon=1.0,
input_dims=(6, ),
n_actions=15,
batch_size=32,
file_name='AI_takeoff/saved_models/dq_model_2.h5')
scores = []
total_steps = []
eps_hist = []
agent.load_model()
for i in range(n_games):
try:
done = False
score = 0
observation = gym_env.reset()
time.sleep(2)
observation_checkpoints = np.array([observation[0:2]])
step_counter = 0
print("GAME ITERATION ", i)
while not done:
action = agent.choose_action(observation)
new_observation, reward, done = gym_env.step(action)
step_counter = step_counter + 1
score = score + reward
agent.store_transition(observation, action, reward,
new_observation, done)
observation = new_observation
# agent.learn()
# This if statement checks if the airplane is stuck
observation_checkpoints = np.append(observation_checkpoints,
[new_observation[0:2]],
axis=0)
print(observation_checkpoints)
print("stepcounter is", step_counter)
if step_counter % 30 == 0:
if np.array_equal(
observation_checkpoints[step_counter - 30],
observation_checkpoints[step_counter - 1]):
done = True
eps_hist.append(agent.epsilon)
scores.append(score)
total_steps.append(step_counter)
except Exception as e:
print(str(e))
def main(env_name=None):
ENV_NAME = 'wumpus-v0'
if env_name: ENV_NAME = env_name
MODEL_DIR = f'models/{ENV_NAME}-dqn'
MODEL_FILE = f'{ENV_NAME}-dqn.h5'
CHECKPOINTS_DIR = f'models/{ENV_NAME}-dqn/checkpoints'
TEST_IMG_DIR = f'tests/{ENV_NAME}-dqn'
env = gym.make(ENV_NAME)
env.reset()
agent = Agent(learning_rate=0.01, gamma=0.95,
state_shape=env.observation_space.shape, actions=7,
batch_size=64,
epsilon_initial=0.0, epsilon_decay=0, epsilon_final=0.0,
replay_buffer_capacity=1000000,
model_name=MODEL_FILE, model_dir=MODEL_DIR,
ckpt_dir=CHECKPOINTS_DIR)
agent.load_model()
done = False
score = 0
steps_per_episode = 0
state = env.reset()
images = [env.render('rgb_array')]
while not done:
# Choose action according to policy, and execute
action = agent.select_action(state)
state, reward, done, _ = env.step(action)
score += reward
steps_per_episode += 1
images.append(env.render('rgb_array'))
# Generate GIF for the execution
create_gif(
f'{ENV_NAME}.gif',
np.array(images),
fps=1.0
)
print(
f'Model \'{str(ENV_NAME)}\', score {score}, steps {steps_per_episode}')
epsilon=0,
epsilon_dec=0.9996,
epsilon_min=0.01,
input_shape=input_shape,
h1_dims=h1_dims,
action_space=action_space,
fname=file_2)
if train_networks == True:
memory_2 = training_tools.ReplayBuffer(mem_size, input_shape,
len(action_space))
# Load networks if specified
if load_networks == True:
if p1_type == 'Agent':
agent_1.load_model(file_1)
if p2_type == 'Agent':
agent_2.load_model(file_2)
print('\n... Models Loaded ...\n')
# Initialize Pong environment
screen_Size = (1000, 600)
env = pongGame(screen_Size, p1_type, p2_type, action_space)
if show_game == True:
framerate = 60
env.setupWindow(framerate)