def run(double_dqn, task_num):
title = "Double_Deep_Q" if double_dqn else "Deep_Q"
env = Monitor(CustomGym(
agentXY,
goalXY,
tasks[task_num][0],
tasks[task_num][1],
title=f"{title}_Task_{task_num + 1}",
),
filename=None)
model = DQN(MlpPolicy,
env,
verbose=verbose,
gamma=gamma,
learning_rate=learning_rate,
double_q=bool(double_dqn))
while len(env.get_episode_rewards()) < episodes:
model.learn(total_timesteps=time_steps)
env.save_csv()
env.destroy()
try:
print('Training Started')
if cfg.POLICY.NAME == 'DDPG':
model.learn(total_timesteps=args.num_timesteps,
log_interval=args.log_interval,
save_path=save_path)
else:
model.learn(total_timesteps=args.num_timesteps,
log_interval=args.log_interval)
finally:
print(100 * '*')
print('FINISHED TRAINING; saving model...')
print(100 * '*')
# save model even if training fails because of an error
model.save(model_dir)
env.destroy()
print('model has been saved.')
# --------------------------------------------------------------------------------------------------------------------"""
# ------------------------------------------------Test----------------------------------------------------------------"""
# --------------------------------------------------------------------------------------------------------------------"""
else: # test
if args.agent_id is not None:
save_path = 'logs/agent_{}/models/'.format(args.agent_id)
else:
save_path = 'logs/'
if args.test_model == '':
best_last = 'best'
if args.test_last: