def main():
env, _ = distdeepq.make_env("Pong")
model = distdeepq.models.cnn_to_dist_mlp(convs=[(32, 8, 4), (64, 4, 2),
(64, 3, 1)],
hiddens=[256],
dueling=False)
act = distdeepq.learn(env,
p_dist_func=model,
lr=1e-4,
max_timesteps=2000000,
buffer_size=10000,
exploration_fraction=0.1,
exploration_final_eps=0.01,
train_freq=4,
learning_starts=10000,
target_network_update_freq=1000,
gamma=0.99,
prioritized_replay=False,
dist_params={
'Vmin': -10,
'Vmax': 10,
'nb_atoms': 51
})
act.save("pong_model.pkl")
env.close()
def main():
env = gym.make("CartPole-v0")
env.seed(1337)
np.random.seed(1337)
tf.set_random_seed(1337)
model = distdeepq.models.dist_mlp([64])
act = distdeepq.learn(env,
p_dist_func=model,
lr=3e-4,
max_timesteps=100000,
buffer_size=50000,
exploration_fraction=0.1,
exploration_final_eps=0.02,
print_freq=10,
callback=callback,
target_network_update_freq=500,
batch_size=32,
gamma=0.95,
dist_params={
'Vmin': 0,
'Vmax': 25,
'nb_atoms': 11
})
print("Saving model to cartpole_model.pkl")
act.save("cartpole_model.pkl")
def exp(env_name='CarRacing-v0',
lr=1e-4,
eps=0.0003125,
max_timesteps=25e6,
buffer_size=1e6,
batch_size=32,
exp_t1=1e6,
exp_p1=0.1,
exp_t2=25e6,
exp_p2=0.01,
train_freq=4,
learning_starts=5e4,
target_network_update_freq=1e4,
gamma=0.99,
num_cpu=50,
nb_atoms=5,
convs=[(32, 8, 4), (64, 4, 2), (64, 3, 1)],
hiddens=[512],
action_res=None):
env = gym.make(env_name)
env = wrapper_car_racing(env) # frame stack
# logger.configure(dir=os.path.join('.', datetime.datetime.now().strftime("openai-%Y-%m-%d-%H-%M-%S-%f")))
logger.configure()
n_action, action_map = get_action_information(env,
env_name,
action_res=action_res)
model = distdeepq.models.cnn_to_dist_mlp(
convs=convs, # [(32, 8, 4), (64, 4, 2), (64, 3, 1)],
hiddens=hiddens, # [512],
# n_action=n_action,
dueling=False)
act = distdeepq.learn(
env,
p_dist_func=model,
lr=lr, # 1e-4
eps=eps,
max_timesteps=int(max_timesteps), # 25M
buffer_size=int(buffer_size), # 1M
batch_size=int(batch_size),
exp_t1=exp_t1,
exp_p1=exp_p1,
exp_t2=exp_t2,
exp_p2=exp_p2,
train_freq=train_freq,
learning_starts=learning_starts, # 50000
target_network_update_freq=target_network_update_freq, # 10000
gamma=gamma,
num_cpu=num_cpu,
prioritized_replay=False,
dist_params={'nb_atoms': nb_atoms},
n_action=int(n_action),
action_map=action_map)
act.save("car_racing_model.pkl")
env.close()
def exp(env_name='Assault', # default setting on server
lr=2.5e-4,
eps=0.0003125,
max_timesteps=25e6,
buffer_size=1e6,
batch_size=32,
exp_t1=1e6,
exp_p1=0.1,
exp_t2=25e6,
exp_p2=0.01,
train_freq=4,
learning_starts=5e4,
target_network_update_freq=1e4,
gamma=0.99,
num_cpu=50,
nb_atoms=5,
convs=[(32, 8, 4), (64, 4, 2), (64, 3, 1)],
hiddens=[512],
):
env, _ = distdeepq.make_env(env_name)
# logging directory setting:
# logger.configure(dir=os.path.join('.', datetime.datetime.now().strftime("openai-%Y-%m-%d-%H-%M-%S-%f")))
# logging configuration: baselines/baselines/logger.py
logger.configure()
model = distdeepq.models.cnn_to_dist_mlp(
convs=convs, #[(32, 8, 4), (64, 4, 2), (64, 3, 1)],
hiddens=hiddens, #[512], #512
dueling=False
)
act = distdeepq.learn(
env,
p_dist_func=model,
lr=lr, # 1e-4
eps=eps,
max_timesteps=int(max_timesteps), # 25M
buffer_size=int(buffer_size), # 1M
batch_size=int(batch_size),
exp_t1=exp_t1,
exp_p1=exp_p1,
exp_t2=exp_t2,
exp_p2=exp_p2,
train_freq=train_freq,
learning_starts=learning_starts, # 50000
target_network_update_freq=target_network_update_freq, # 10000
gamma=gamma,
num_cpu=num_cpu,
prioritized_replay=False,
dist_params={'nb_atoms': nb_atoms}
)
act.save("assault_model.pkl")
env.close()