0

执行“3Dball”会在 Unity ml-agent 中产生一些错误

当我执行 PPO.ipynb 时，直到“加载环境”才出现错误。

执行“训练代理”有一些错误

ValueError：列出参数“值”到“ConcatV2”操作，长度 0 比最小长度 2 短。

这是我执行的代码

https://github.com/Unity-Technologies/ml-agents/blob/master/python/PPO.ipynb

tf.reset_default_graph()

if curriculum_file == "None":
    curriculum_file = None


def get_progress():
    if curriculum_file is not None:
        if env._curriculum.measure_type == "progress":
            return steps / max_steps
        elif env._curriculum.measure_type == "reward":
            return last_reward
        else:
            return None
    else:
        return None

# Create the Tensorflow model graph
ppo_model = create_agent_model(env, lr=learning_rate,
                               h_size=hidden_units, epsilon=epsilon,
                               beta=beta, max_step=max_steps, 
                               normalize=normalize, num_layers=num_layers)

is_continuous = (env.brains[brain_name].action_space_type == "continuous")
use_observations = (env.brains[brain_name].number_observations > 0)
use_states = (env.brains[brain_name].state_space_size > 0)

model_path = './models/{}'.format(run_path)
summary_path = './summaries/{}'.format(run_path)

if not os.path.exists(model_path):
    os.makedirs(model_path)

if not os.path.exists(summary_path):
    os.makedirs(summary_path)

init = tf.global_variables_initializer()
saver = tf.train.Saver()

with tf.Session() as sess:
    # Instantiate model parameters
    if load_model:
        print('Loading Model...')
        ckpt = tf.train.get_checkpoint_state(model_path)
        saver.restore(sess, ckpt.model_checkpoint_path)
    else:
        sess.run(init)
    steps, last_reward = sess.run([ppo_model.global_step, ppo_model.last_reward])    
    summary_writer = tf.summary.FileWriter(summary_path)
    info = env.reset(train_mode=train_model, progress=get_progress())[brain_name]
    trainer = Trainer(ppo_model, sess, info, is_continuous, use_observations, use_states, train_model)
    if train_model:
        trainer.write_text(summary_writer, 'Hyperparameters', hyperparameter_dict, steps)
    while steps <= max_steps:
        if env.global_done:
            info = env.reset(train_mode=train_model, progress=get_progress())[brain_name]
        # Decide and take an action
        new_info = trainer.take_action(info, env, brain_name, steps, normalize)
        info = new_info
        trainer.process_experiences(info, time_horizon, gamma, lambd)
        if len(trainer.training_buffer['actions']) > buffer_size and train_model:
            # Perform gradient descent with experience buffer
            trainer.update_model(batch_size, num_epoch)
        if steps % summary_freq == 0 and steps != 0 and train_model:
            # Write training statistics to tensorboard.
            trainer.write_summary(summary_writer, steps, env._curriculum.lesson_number)
        if steps % save_freq == 0 and steps != 0 and train_model:
            # Save Tensorflow model
            save_model(sess, model_path=model_path, steps=steps, saver=saver)
        steps += 1
        sess.run(ppo_model.increment_step)
        if len(trainer.stats['cumulative_reward']) > 0:
            mean_reward = np.mean(trainer.stats['cumulative_reward'])
            sess.run(ppo_model.update_reward, feed_dict={ppo_model.new_reward: mean_reward})
            last_reward = sess.run(ppo_model.last_reward)
    # Final save Tensorflow model
    if steps != 0 and train_model:
        save_model(sess, model_path=model_path, steps=steps, saver=saver)
env.close()
export_graph(model_path, env_name)
4

1 回答 1

0

我遇到了同样的错误，我修复它的方法是替换文件下的第 222 行：“ml-agents/python/ppo/models.py”：

替换第 222 行：

    hidden_visual = tf.concat(encoders, axis=2)

经过：

    if encoders:
        hidden_visual = tf.concat(encoders, axis=2)

我希望这对你有所帮助。

于 2018-01-22T12:08:19.943 回答