def agent():
env = RemoteEnv(IN_PORT=8080, OUT_PORT=8081, host="127.0.0.1")
env.open(0)
actions = {'w': 0, 's': 3, 'a': 4, 'd': 1}
for i in range(10000000):
#state = env.step("restart")
state = env.step("restart")
prev_energy = state['energy']
done = state['done']
print("OK")
while not done:
print('Action: ', end='')
action = getch.getche()
#action = np.random.choice([0, 1])
reward_sum = 0
state = env.step("act", actions[action])
reward_sum += state['reward']
for i in range(1):
done = state['done']
if (not done):
state = env.step('act', -1)
reward_sum += state['reward']
else:
break
energy = state['energy']
prev_energy = energy
frame = state['frame']
print(transform(frame))
print('Reward: ', reward_sum)
print('Done: ', state['done'])
prev_energy = energy
if i >= 2:
break
env.close()
def agent():
env = RemoteEnv(IN_PORT=8080, OUT_PORT=7070, host="127.0.0.1")
env.open(0)
for i in range(10000000):
#state = env.step("restart")
state = env.step("restart")
prev_energy = state['energy']
done = state['done']
print("OK")
while not done:
action = int(input('action'))
#action = np.random.choice([0, 1])
reward_sum = 0
touched = np.zeros(8)
for i in range(8):
state = env.step("act", action)
energy = state['energy']
touched[i] = state['touched']
reward_sum += state['reward']
if state['done']:
break
done = state['done']
prev_energy = energy
frame = state['frame']
frame = image_decode(state['frame'], 20, 20)
print(frame)
print('Reward: ', reward_sum)
print('Touched: ', touched)
print('Signal: ', state['signal'])
print('Done: ', state['done'])
prev_energy = energy
if i >= 2:
break
env.close()
def agent():
env = RemoteEnv(IN_PORT=8081, OUT_PORT=7071)
env.open(0)
for i in range(10000000):
state = env.step("restart")
done = state['done']
print('new episode')
while not done:
frame = image_from_str(state['state'], 10, 10)
print(frame)
action = int(input('action'))
state = env.step("move", action)
done = state['done']
print(state)
env.close()
def agent():
env = RemoteEnv(IN_PORT=8080, OUT_PORT=7070, host="127.0.0.1")
env.open(0)
actions = {
'w': ('walk', 10),
's': ('walk', -10),
'd': ('walk_in_circle', 10),
'a': ('walk_in_circle', -10),
'r': ('run', 20)
}
for i in range(10000000):
#state = env.step("restart")
state = env.step("restart")
prev_energy = state['energy']
done = state['done']
print("OK")
while not done:
print('Action: ', end='')
action = getch.getche()
#action = np.random.choice([0, 1])
reward_sum = 0
for i in range(7):
state = env.step(actions[action][0], actions[action][1])
reward_sum += state['reward']
if (done):
break
if not done:
state = env.step("get_result")
reward_sum += state['reward']
done = state['done']
energy = state['energy']
prev_energy = energy
frame = state['frame']
print(transform(frame))
print('Reward: ', reward_sum)
print('Done: ', state['done'])
print('Delta energy: ', prev_energy - energy)
prev_energy = energy
if i >= 2:
break
env.close()
from ai4u.core import RemoteEnv
env = RemoteEnv()
env.open()
env.step('tx', 5)
env.close()
from ai4u.core import RemoteEnv
import numpy as np
env = RemoteEnv()
env.open(0)
if __name__ == "__main__":
for i in range(10):
sum_reward = 0
state = env.step("restart")
print(state)
done = state['done']
while not done:
frame = None
action = int(
input("action-----------------------------------------"))
if action == 0:
state = env.step("fx", 0.1)
elif action == 1:
state = env.step("fz", 0.1)
else:
state = env.step("noop", 0.0)
done = state["done"]
reward = state['reward']
sum_reward += reward
print(sum_reward)
env.close()
def configure(self, environment_definitions, port_inc=0):
if 'action_space' in environment_definitions:
self.action_space = environment_definitions['action_space']
elif 'action_shape' in environment_definitions:
self.action_space = spaces.Discrete(
environment_definitions['action_shape'][0])
else:
raise Exception(
'environment_definitions do not contains action_space or action_shape'
)
assert 'actions' in environment_definitions, 'environment_definitions do not contain actions (a list of actions descriptors)'
self.actions = environment_definitions['actions']
if 'observation_space' in environment_definitions:
self.observation_space = environment_definitions[
'observation_space']
elif 'state_shape' in environment_definitions:
state_shape = environment_definitions['state_shape']
state_type = environment_definitions['state_type']
min_value = environment_definitions['min_value']
max_value = environment_definitions['max_value']
self.observation_space = spaces.Box(low=min_value,
high=max_value,
shape=state_shape,
dtype=state_type)
else:
raise Exception(
'environment_definitions do not contain observation_space or state_shape.'
)
if 'n_envs' in environment_definitions:
self.n_envs = environment_definitions['n_envs']
else:
self.n_envs = 1
self.verbose = False
if 'verbose' in environment_definitions:
self.verbose = environment_definitions['verbose']
if 'action_meaning' in environment_definitions:
self.action_meaning = environment_definitions['action_meaning']
else:
self.action_meaning = ['action'] * len(self.actions)
if 'seed' in environment_definitions:
self._seed = environment_definitions['seed']
else:
self._seed = 0
if 'agent' in environment_definitions:
if inspect.isclass(environment_definitions['agent']):
self.agent = environment_definitions['agent']()
else:
raise Exception("Agent object is not a class!!!!")
else:
self.agent = BasicAgent()
host = '127.0.0.1'
if host in environment_definitions:
host = environment_definitions['host']
input_port = 8080
if 'input_port' in environment_definitions:
input_port = environment_definitions['input_port']
output_port = 8081
if 'output_port' in environment_definitions:
output_port = environment_definitions['output_port']
self.remoteenv = RemoteEnv(host, output_port + port_inc,
input_port + port_inc)
self.id = port_inc
self.remoteenv.verbose = self.verbose
self.remoteenv.open(0)
self.nlives = 1
self.state = None
self.configureFlag = True
class Environment(gym.Env):
metadata = {'render.modes': ['human']}
def __init__(self):
super(Environment, self).__init__()
self.ale = AleWrapper(self)
self.configureFlag = False
self.id = 0
if not (BasicAgent.environment_definitions is None):
self.configure(BasicAgent.environment_definitions,
BasicAgent.environment_port_id)
BasicAgent.environment_port_id += 1
def configure(self, environment_definitions, port_inc=0):
if 'action_space' in environment_definitions:
self.action_space = environment_definitions['action_space']
elif 'action_shape' in environment_definitions:
self.action_space = spaces.Discrete(
environment_definitions['action_shape'][0])
else:
raise Exception(
'environment_definitions do not contains action_space or action_shape'
)
assert 'actions' in environment_definitions, 'environment_definitions do not contain actions (a list of actions descriptors)'
self.actions = environment_definitions['actions']
if 'observation_space' in environment_definitions:
self.observation_space = environment_definitions[
'observation_space']
elif 'state_shape' in environment_definitions:
state_shape = environment_definitions['state_shape']
state_type = environment_definitions['state_type']
min_value = environment_definitions['min_value']
max_value = environment_definitions['max_value']
self.observation_space = spaces.Box(low=min_value,
high=max_value,
shape=state_shape,
dtype=state_type)
else:
raise Exception(
'environment_definitions do not contain observation_space or state_shape.'
)
if 'n_envs' in environment_definitions:
self.n_envs = environment_definitions['n_envs']
else:
self.n_envs = 1
self.verbose = False
if 'verbose' in environment_definitions:
self.verbose = environment_definitions['verbose']
if 'action_meaning' in environment_definitions:
self.action_meaning = environment_definitions['action_meaning']
else:
self.action_meaning = ['action'] * len(self.actions)
if 'seed' in environment_definitions:
self._seed = environment_definitions['seed']
else:
self._seed = 0
if 'agent' in environment_definitions:
if inspect.isclass(environment_definitions['agent']):
self.agent = environment_definitions['agent']()
else:
raise Exception("Agent object is not a class!!!!")
else:
self.agent = BasicAgent()
host = '127.0.0.1'
if host in environment_definitions:
host = environment_definitions['host']
input_port = 8080
if 'input_port' in environment_definitions:
input_port = environment_definitions['input_port']
output_port = 8081
if 'output_port' in environment_definitions:
output_port = environment_definitions['output_port']
self.remoteenv = RemoteEnv(host, output_port + port_inc,
input_port + port_inc)
self.id = port_inc
self.remoteenv.verbose = self.verbose
self.remoteenv.open(0)
self.nlives = 1
self.state = None
self.configureFlag = True
def seed(self, seed=None):
self.agent.seed(seed, self)
def get_action_meanings(self):
self.__check_configuration_()
return [self.action_meaning[i] for i in range(len(self.actions))]
def reset(self):
self.__check_configuration_()
return self.agent.reset(self)
def render(self, mode='human', close=False):
self.__check_configuration_()
self.agent.render()
def close(self):
self.__check_configuration_()
self.remoteenv.close()
def one_step(self, action):
return self.remoteenv.step(self.actions[action][0],
self.actions[action][1])
def one_stepfv(self, action):
return self.remoteenv.stepfv(self.actions[action][0],
self.actions[action][1])
def step(self, action, info=None):
self.__check_configuration_()
return self.agent.act(self, action, info)
def __del__(self):
self.__check_configuration_()
self.remoteenv.close()
def __check_configuration_(self):
if not self.configureFlag:
raise Exception(
"The environment is not configured. Try to set up the environment before trying again!!!"
)