if train_model:
trainer.write_text(summary_writer, 'Hyperparameters', options, steps)
while steps <= max_steps or not train_model:
if env.global_done:
info = env.reset(train_mode=train_model, progress=get_progress())[brain_name]
trainer.reset_buffers(info, total=True)
# 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)
if train_model:
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()
graph_name = (env_name.strip()
sess.run(init)
steps = sess.run(ppo_model.global_step)
summary_writer = tf.summary.FileWriter(summary_path)
info = env.reset(train_mode=train_model)[brain_name]
trainer = Trainer(ppo_model, sess, info, is_continuous, use_observations,
use_states)
while steps <= max_steps or not train_model:
if env.global_done:
info = env.reset(train_mode=train_model)[brain_name]
# Decide and take an action
new_info = trainer.take_action(info, env, brain_name)
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)
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)
# 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)
trainer.reset_buffers(info, total=True)
# Decide and take an action
info = trainer.take_action(info,
env,
brain_name,
steps,
normalize,
stochastic=True)
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, episode_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)
if train_model:
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)
if not watcher_started and render:
watcher = threading.Thread(target=watch,
args=(tf.get_default_session(), ))
watcher.start()