score_book = {}
value_loss_book = {}
actor_loss_book = {}
critic_1_loss_book = {}
critic_2_loss_book = {}
critic_loss_book = {}
for trial_num in range(n_trials):
print('\nTrial num:', trial_num + 1)
agent = Agent(input_dims=env.observation_space.shape,
layer1_size=200,
layer2_size=200,
env=env,
n_actions=env.action_space.shape[0],
alpha=alpha,
beta=beta,
gamma=gamma,
max_size=max_size,
tau=tau,
ent_alpha=ent_alpha,
batch_size=batch_size,
reward_scale=reward_scale,
chkpt_dir=chkpt_dir)
score_history = []
loss = []
value_loss = []
actor_loss = []
critic_1_loss = []
critic_2_loss = []
critic_loss = []
if load_checkpoint:
perturbations = [0.3, 0.5, 0.7, 0.9]
for perturbation in perturbations:
reward_history = []
score_book = {}
value_loss_book = {}
actor_loss_book = {}
critic_1_loss_book = {}
critic_2_loss_book = {}
critic_loss_book = {}
print("Perturbation amount: {}".format(perturbation))
for trial_num in range(n_trials):
print('\nTrial num:', trial_num+1)
agent = Agent(input_dims=env.observation_space.shape, layer1_size=256, layer2_size=256,
env=env, n_actions=env.action_space.n, alpha=alpha, beta=beta,
gamma=gamma, max_size=max_size, tau=tau, ent_alpha=ent_alpha, batch_size=batch_size,
reward_scale = reward_scale, chkpt_dir=chkpt_dir)
score_history = []
loss = []
value_loss = []
actor_loss = []
critic_1_loss = []
critic_2_loss = []
critic_loss = []
if load_checkpoint:
agent.load_models()
for i in tqdm(range(n_games)):
if render:
if __name__ == '__main__':
#=========================Enable GPU usage===========================
physical_devices = tf.config.experimental.list_physical_devices('GPU')
tf.config.experimental.set_memory_growth(physical_devices[0], True)
#========Operational Instructions=========
render = True
debugging = False
training = True
load_checkpoint = True
#============================Constants================
n_episodes = 7000
env = CarRacing(obstacles=False)
agent = Agent(env)
total_steps = 0
if load_checkpoint:
total_rewards = list(agent.load_rewards('total'))
avg_rewards = list(agent.load_rewards('avg'))
episode_offset = len(total_rewards)
agent.load_models()
agent.load_buffer()
else:
total_rewards = []
avg_rewards = []
#================Main loop===============================
for episode in range(n_episodes):
if load_checkpoint: episode += episode_offset
env.reset()