def normal_goal():
env = Breakout(conf)
env = BreakoutFullObservableStateWrapper(env)
# observation_space = env.observation_space
# action_space = env.action_space
# feat_ext = BreakoutRobotFeatureExtractor()
# feature_space = feat_ext.output_space
# print(observation_space, action_space, feature_space)
agent = RLAgent(
BreakoutRobotFeatureExtractor(),
RandomPolicy(env.action_space,
epsilon_start=1.0,
epsilon_end=0.01,
decaying_steps=1000000),
Sarsa(None, env.action_space, alpha=None, gamma=0.99, nsteps=100))
return env, agent
def temporal_goal():
env = Breakout(conf)
env = BreakoutFullObservableStateWrapper(env)
# observation_space = env.observation_space
# action_space = env.action_space
# robot_feat_ext = BreakoutRobotFeatureExtractor()
# feature_space = robot_feat_ext.output_space
# print(observation_space, action_space, feature_space)
agent = TGAgent(
BreakoutRobotFeatureExtractor(),
RandomPolicy(env.action_space,
epsilon_start=1.0,
epsilon_end=0.01,
decaying_steps=7500000),
Sarsa(None, env.action_space, alpha=None, gamma=0.99, nsteps=200),
[BreakoutRowBottomUpTemporalEvaluator()])
return env, agent
obs_space.spaces["color"],
obs_space.spaces["RAState"],
))
super().__init__(obs_space, robot_feature_space)
def _extract(self, input, **kwargs):
return (input["x"], input["y"], input["theta"], input["color"],
input["RAState"])
if __name__ == '__main__':
env = GymSapientino()
'''Normal task - no temporal goal'''
agent = RLAgent(
SapientinoNRobotFeatureExtractor(env.observation_space),
RandomPolicy(env.action_space, epsilon=0.1),
QLearning(None, env.action_space, alpha=0.1, gamma=0.9, nsteps=100))
t = Trainer(
env,
agent,
n_episodes=100000,
resume=False,
eval=False,
# resume = True,
# eval = True,
# renderer=PygameRenderer(delay=0.05)
)
t.main()
obs_space.spaces["task_state"],
))
super().__init__(obs_space, robot_feature_space)
def _extract(self, input, **kwargs):
return (input["x"], input["y"], input["theta"], input["task_state"])
if __name__ == '__main__':
env = GymMinecraft()
'''Normal task - no temporal goal'''
agent = RLAgent(
MinecraftNRobotFeatureExtractor(env.observation_space),
RandomPolicy(env.action_space,
epsilon=0.01,
epsilon_start=1.0,
decaying_steps=1),
QLearning(None, env.action_space, alpha=None, gamma=0.9, nsteps=200))
t = Trainer(
env,
agent,
n_episodes=100000,
# resume=False,
# eval=False,
resume=True,
eval=True,
renderer=PygameRenderer(delay=0.01))
t.main()
import gym
from rltg.agents.RLAgent import RLAgent
from rltg.agents.brains.TDBrain import Sarsa, QLearning
from rltg.agents.exploration_policies.RandomPolicy import RandomPolicy
from rltg.agents.feature_extraction import IdentityFeatureExtractor
from rltg.trainer import Trainer
from rltg.utils.GoalEnvWrapper import GoalEnvWrapper
if __name__ == '__main__':
env = gym.make("Taxi-v2")
env = GoalEnvWrapper(env, lambda _, reward, done, __: done and reward==20)
observation_space = env.observation_space
action_space = env.action_space
print(observation_space, action_space)
agent = RLAgent(
IdentityFeatureExtractor(observation_space),
RandomPolicy(action_space, epsilon=0.1, epsilon_start=1.0, decaying_steps=5000),
QLearning(observation_space, action_space, alpha=0.8, nsteps=1, gamma=1.0)
)
tr = Trainer(env, agent, n_episodes=10000, resume=False, eval=False, window_size=1000)
# tr = Trainer(env, agent, n_episodes=10000, resume=True, eval=True, window_size=1000)
tr.main()