def test_linear_sarsa(iterations=1000, mlambda=None, n0=100, avg_it=100):
print "\n-------------------"
print "TD control Sarsa, with Linear function approximation"
print "run for n. iterations: "+str(iterations)
print "plot graph mse vs episodes for lambda equal 0 and lambda equal 1"
print "list (std output) win percentage for values of lambda 0, 0.1, 0.2, ..., 0.9, 1"
monte_carlo_Q = pickle.load(open("Data/Qval_func_1000000_MC_control.pkl", "rb"))
n_elements = monte_carlo_Q.shape[0]*monte_carlo_Q.shape[1]*2
mse = []
if not isinstance(mlambda,list):
# if no value is passed for lambda, default 0.5
l = 0.5 if mlambda==None else mlambda
# learn
game = Environment()
agent = Agent(game, n0)
agent.TD_control_linear(iterations,l,avg_it)
agent.show_statevalue_function()
else:
# test each value of lambda
for l in mlambda:
game = Environment()
agent = Agent(game, n0)
l_mse = agent.TD_control_linear(iterations,l,avg_it)
mse.append(l_mse)
plt.plot(mlambda,mse)
plt.ylabel('mse')
plt.show()
def test_monte_carlo(iterations=1000000, n0=100):
print "\n-------------------"
print "Monte Carlo control"
print "run for n. iterations: "+str(iterations)
print "win percentage: "
# learn
game = Environment()
agent = Agent(game, n0)
agent.MC_control(iterations)
# plot and store
agent.show_statevalue_function()
agent.store_Qvalue_function()
def test_linear_sarsa(iterations=1000, mlambda=None, n0=100, avg_it=100):
print "\n-------------------"
print "TD control Sarsa, with Linear function approximation"
print "run for n. iterations: " + str(iterations)
print "plot graph mse vs episodes for lambda equal 0 and lambda equal 1"
print "list (std output) win percentage for values of lambda 0, 0.1, 0.2, ..., 0.9, 1"
monte_carlo_Q = pickle.load(
open("Data/Qval_func_1000000_MC_control.pkl", "rb"))
n_elements = monte_carlo_Q.shape[0] * monte_carlo_Q.shape[1] * 2
mse = []
if not isinstance(mlambda, list):
# if no value is passed for lambda, default 0.5
l = 0.5 if mlambda == None else mlambda
# learn
game = Environment()
agent = Agent(game, n0)
agent.TD_control_linear(iterations, l, avg_it)
agent.show_statevalue_function()
else:
# test each value of lambda
for l in mlambda:
game = Environment()
agent = Agent(game, n0)
l_mse = agent.TD_control_linear(iterations, l, avg_it)
mse.append(l_mse)
plt.plot(mlambda, mse)
plt.ylabel('mse')
plt.show()
def reset(self, training=True):
# Generate new obstacles
self.generate_obstacles(self.n_rows, self.n_cols)
# Generate agents positions
team_agents, team_opponents = self.generate_random_agents(
n_agents=self.n_agents,
n_opponents=self.n_opponents,
rows=self.n_rows,
cols=self.n_cols,
cluster=True)
# Instantiate agents
self.agents = [Agent(pos, i == 0) for i, pos in enumerate(team_agents)]
self.opponents = [DummyAgent(position=pos) for pos in team_opponents]
# Assign training and target nets to agents
for a in self.agents:
if a.training:
a.brain = self.training_net
else:
a.brain = self.target_net
# Generate all allowed moves for each position of the board
self.allowed_moves_per_position = self.create_allowed_moves()
self.training_net.training = training
def test_presence_request_server_response():
sensor1 = Sensor('A222', 21.02, 'second', 'accelerometr')
sensor2 = Sensor('B333', 56.0003, 'meter', 'mleko')
agent = Agent(1, 'xxx', 'rrr', 'yyy', [sensor1, sensor2])
response = log.presence_request(agent, 'mleko', None, None, 'SL', None,
None)
assert response.status != 404 and response.status != 500
def test__presence_request_json():
agent = Agent(1, 'Adam1', 'Testowy agent', 'agent_mobilny', [])
result = log._presence_request_json(agent,
res('logger\\precence_type\\start'),
protocol=None,
request_type=None,
language='SL',
action_receiver_id=3,
action_sender_id=None)
json_data = json.loads(result)
assert json_data['content'] == 'presenceStatus:start_presence,agent:Adam1'
def test_monte_carlo(iterations=1000000, n0=100):
print "\n-------------------"
print "Monte Carlo control"
print "run for n. iterations: " + str(iterations)
print "win percentage: "
# learn
game = Environment()
agent = Agent(game, n0)
agent.MC_control(iterations)
# plot and store
agent.show_statevalue_function()
agent.store_Qvalue_function()
def test__agent_registration_json_agent_without_sensors():
agent = Agent(1, 'Adam1', 'Testowy agent', 'agent_mobilny', [])
result = log._agent_registration_json(agent,
protocol=None,
request_type=None,
language='SL',
action_receiver_id=3,
action_sender_id=None)
json_data = json.loads(result)
assert json_data[
'content'] == 'agentName:Adam1,agentDescription:Testowy agent,agentType:agent_mobilny,sensors:'
assert json_data['action_protocol'] is None
assert json_data['action_requesttype'] is None
assert json_data['language'] == 'SL'
assert json_data['action_receiver']['id'] == 3
assert json_data['action_sender'] is None
def __init__(self,
n_rows,
n_cols,
n_agents,
n_opponents,
gcr_weight=0.5,
lcr_weight=0.5,
rr_weight=0.5):
self.n_rows = n_rows
self.n_cols = n_cols
self.n_agents = n_agents
self.n_opponents = n_opponents
self.obstacles = None
self.allowed_moves_per_position = None
self.generate_obstacles(n_rows, n_cols)
team_agents, team_opponents = self.generate_random_agents(
n_agents=n_agents,
n_opponents=n_opponents,
rows=n_rows,
cols=n_cols,
cluster=True)
self.agents = [Agent(pos, i == 0) for i, pos in enumerate(team_agents)]
self.training_net = Brain(True)
self.target_net = Brain(False)
for a in self.agents:
if a.training:
a.brain = self.training_net
else:
a.brain = self.target_net
self.opponents = [DummyAgent(position=pos) for pos in team_opponents]
self.global_capture_weight = gcr_weight
self.local_capture_weight = gcr_weight
self.reachability_weight = rr_weight
self.training_ate_weight = lcr_weight
def test__agent_registration_json_agent_with_sensors():
agent = Agent(1, 'Adam1', 'Testowy agent', 'agent_mobilny', [])
sensor1 = Sensor('A222', 21.02, 'second', 'accelerometr')
sensor2 = Sensor('B333', 56.0003, 'meter', 'mleko')
agent.sensor_list.append(sensor1)
agent.sensor_list.append(sensor2)
result = log._agent_registration_json(agent,
protocol=None,
request_type=None,
language='SL',
action_receiver_id=None,
action_sender_id=1)
json_data = json.loads(result)
assert json_data['content'] == 'agentName:Adam1,agentDescription:Testowy agent,agentType:agent_mobilny,' \
+ 'sensors:sensorName-A222 sensorAccuracy-21.02 ' \
+ 'unitType-second sensorType-accelerometr;' \
+ 'sensorName-B333 sensorAccuracy-56.0003 unitType-meter sensorType-mleko'
assert json_data['action_protocol'] is None
assert json_data['action_requesttype'] is None
assert json_data['language'] == 'SL'
assert json_data['action_receiver'] is None
assert json_data['action_sender']['id'] == 1
def test_agent_registration_server_response_empty_strings():
agent = Agent(1, '', '', '', [])
response = log.agent_registration(agent, None, None, 'SL', None, None)
assert response.status != 404 and response.status != 500
def test_presence_request_server_response_empty_strings():
agent = Agent(1, '', '', '', [])
response = log.presence_request(agent, 'mleko', None, None, 'SL', None,
None)
assert response.status != 404 and response.status != 500
def test_td_control(iter=50000):
lambdas = [0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]
for l in lambdas:
agent = Agent()
agent.TD_control(iter, l)
agent.plot_state()
def test_mc_control(iter=50000, n0=100):
agent = Agent()
agent.MC_control(iter, n0)
agent.plot_state()
def test_td_control_linear_app(iter=50000):
lambdas = [0.0, 1.0]
for l in lambdas:
agent = Agent()
agent.TD_control_linear_app(iter, 0.9)
agent.plot_state()
def get_agents():
return agent
def perform_agent(name, method: str):
getattr(agent, method)()
def perform_recorder(name, method: str):
recorder = [recorder for recorder in recorders if recorder.name == name][0]
getattr(recorder, method)()
recorders = create_recorders()
agent = Agent("Main Agent", recorders)
# agent.start()
#
# agent.stop()
remote_api_cfg = cfg.get_configuration_of('remote_api_config')
api_server = RemoteAPIServer(remote_api_cfg.get('ip'),
remote_api_cfg.get('port'))
logging.debug("Create RemoteAPIServer")
api_server.register_function(get_agents)
api_server.register_function(perform_agent)
api_server.register_function(perform_recorder)
api_server.start_listen()