def main(env_id, render, num_process, lr_p, lr_v, gamma, tau, epsilon, batch_size,
ppo_mini_batch_size, ppo_epochs, max_iter, eval_iter, save_iter, model_path, log_path, seed):
base_dir = log_path + env_id + "/PPO_exp{}".format(seed)
writer = SummaryWriter(base_dir)
ppo = PPO(env_id=env_id,
render=render,
num_process=num_process,
min_batch_size=batch_size,
lr_p=lr_p,
lr_v=lr_v,
gamma=gamma,
tau=tau,
clip_epsilon=epsilon,
ppo_epochs=ppo_epochs,
ppo_mini_batch_size=ppo_mini_batch_size,
seed=seed)
for i_iter in range(1, max_iter + 1):
ppo.learn(writer, i_iter)
if i_iter % eval_iter == 0:
ppo.eval(i_iter, render=render)
if i_iter % save_iter == 0:
ppo.save(model_path)
pickle.dump(ppo,
open('{}/{}_ppo.p'.format(model_path, env_id), 'wb'))
torch.cuda.empty_cache()
def main(env_id, render, num_process, lr_p, lr_v, gamma, tau, epsilon, batch_size,
ppo_mini_batch_size, ppo_epochs, model_path, seed, test_epochs):
ppo = PPO(env_id=env_id,
render=render,
num_process=num_process,
min_batch_size=batch_size,
lr_p=lr_p,
lr_v=lr_v,
gamma=gamma,
tau=tau,
clip_epsilon=epsilon,
ppo_epochs=ppo_epochs,
ppo_mini_batch_size=ppo_mini_batch_size,
seed=seed,
model_path=model_path)
for i_iter in range(1, test_epochs):
ppo.eval(i_iter)