Howto RL-ATT-001: Train and Reload Single Agent using Stagnation Detection (Gym)
Prerequisites
- Please install the following packages to run this examples properly:
Executable code
## -------------------------------------------------------------------------------------------------
## -- Project : MLPro - A Synoptic Framework for Standardized Machine Learning Tasks
## -- Package : mlpro.rl.examples
## -- Module : howto_rl_att_001_train_and_reload_single_agent_gym_sd.py
## -------------------------------------------------------------------------------------------------
## -- History :
## -- yyyy-mm-dd Ver. Auth. Description
## -- 2023-03-04 1.0.0 DA Creation as derivate of howto_rl_agent_011
## -- 2023-03-27 1.0.1 DA Refactoring
## -------------------------------------------------------------------------------------------------
"""
Ver. 1.0.1 (2023-03-27)
As in Howto RL AGENT 011, this module shows how to train a single agent and load it again to do some
extra cycles. In opposite to howto 011, stagnation detection is used to automatically end the
training if no further progress can be made.
You will learn:
1. How to use the RLScenario class of MLPro.
2. How to save a scenario after some run.
3. How to reload the saved scenario and re-run for additional cycles.
4. How to use stagnation detection to end the training automatically if there is no progress.
"""
import gym
from stable_baselines3 import PPO
from mlpro.rl import *
from mlpro.wrappers.openai_gym import WrEnvGYM2MLPro
from mlpro.wrappers.sb3 import WrPolicySB32MLPro
from pathlib import Path
# 1 Implement your own RL scenario
class MyScenario (RLScenario):
C_NAME = 'Matrix'
def _setup(self, p_mode, p_ada: bool, p_visualize: bool, p_logging) -> Model:
# 1.1 Setup environment
gym_env = gym.make('CartPole-v1')
self._env = WrEnvGYM2MLPro(gym_env, p_visualize=p_visualize, p_logging=p_logging)
# 1.2 Setup Policy From SB3
policy_sb3 = PPO(
policy="MlpPolicy",
n_steps=10,
env=None,
_init_setup_model=False,
device="cpu",
seed=1)
# 1.3 Wrap the policy
policy_wrapped = WrPolicySB32MLPro(
p_sb3_policy=policy_sb3,
p_cycle_limit=self._cycle_limit,
p_observation_space=self._env.get_state_space(),
p_action_space=self._env.get_action_space(),
p_ada=p_ada,
p_visualize=p_visualize,
p_logging=p_logging)
# 1.4 Setup standard single-agent with own policy
return Agent(
p_policy=policy_wrapped,
p_envmodel=None,
p_name='Smith',
p_ada=p_ada,
p_visualize=p_visualize,
p_logging=p_logging
)
if __name__ == '__main__':
# Parameters for demo mode
cycle_limit = 20000
adaptation_limit = 0
stagnation_limit = 5
eval_frequency = 10
eval_grp_size = 5
logging = Log.C_LOG_WE
visualize = True
path = str(Path.home())
else:
# Parameters for internal unit test
cycle_limit = 50
adaptation_limit = 5
stagnation_limit = 5
eval_frequency = 2
eval_grp_size = 1
logging = Log.C_LOG_NOTHING
visualize = False
path = str(Path.home())
# 2 Create scenario and start training
training = RLTraining(
p_scenario_cls=MyScenario,
p_cycle_limit=cycle_limit,
p_adaptation_limit=adaptation_limit,
p_stagnation_limit=stagnation_limit,
p_eval_frequency=eval_frequency,
p_eval_grp_size=eval_grp_size,
p_path=path,
p_visualize=visualize,
p_logging=logging )
# 3 Training
training.run()
filename_scenario = training.get_scenario().get_filename()
# 4 Reload the scenario
if __name__ == '__main__':
input( '\nTraining finished. Press ENTER to reload and run the scenario...\n')
scenario = MyScenario.load( p_path = training.get_training_path() + os.sep + 'scenario',
p_filename = filename_scenario )
# 5 Reset Scenario
scenario.reset()
# 6 Run Scenario
scenario.run()
if __name__ != '__main__':
from shutil import rmtree
rmtree(training.get_training_path())
else:
input( '\nPress ENTER to finish...')
Results
The Gym Cartpole environment window appears during training and shows an improved control behavior after a while. After the training, the related scenario is reloaded and run for a further episode to demonstrate the final control behavior.
The training itself is terminated due to automatic stagnation detection. The chart below shows the training progress and the ending at the point of maximum possible reward:
- After termination the local result folder contains the training result files:
agent_actions.csv
env_rewards.csv
env_states.csv
evaluation.csv
summary.csv
scenario
Cross Reference