def __init__(self, torcs_client):
self.torcs = torcs_client
self.population = []
self.previous_population = []
self.fitness_history = []
self.pop_fitness_history = []
self.track = 'random'
self.tracks_used = []
self.laps = 3
self.output_file = trained_files.get_file("best_evo_driver.pkl")
# set this to true to have fitness values comparable across different tracks
self.normalize_fitness=False
def get_action(self, sensors):
try:
x = sensors.get_array(self.sensor_ids)
if self.w_in is None and self.w_back is None and self.w_out is None:
self.random_weights(n_input=len(x)+1)
u = np.matrix(np.concatenate(([self.bias], x), axis=0))
x_n_update = self.activation(self.w_in.dot(u.T) + self.w.dot(self.x_n) + self.w_back.dot(self.y))
self.x_n = self.leak_rate * x_n_update + (1 - self.leak_rate) * self.x_n
self.y = np.squeeze(np.array(self.activation(self.w_out.T.dot(self.x_n)))) * 2 - 1.0
return self.y
except Exception as e:
print "ESN Error:", e
return 0.0
if __name__ == '__main__':
import sys
import os
sys.path.insert(0, os.path.dirname(os.path.realpath('network.py')) + '/../')
import logs
import trained_files
from client import SENSOR_IDXS
steer = FFNetwork(sensor_ids=SENSOR_IDXS, action_ids=['steering'], n_hidden=5)
training_files = [logs.get_file("lau/"+f) for f in os.listdir(logs.get_file("lau/"))][:5]
print training_files
steer.train(training_files=training_files, scaling=False)
with open(trained_files.get_file("pybrain_test.pkl"), 'wb') as f:
pickle.dump(steer, f)
sys.exit(1)
c = Client(torcs_path=sys.argv[1])
# CONFIGURATION #
POPULATION_SIZE = 10
BATCHES = 5
EVO_ITERATIONS = 3
TRACK = 'random'
LAPS = 2
NORMALIZED_FITNESS = True
c.timeout = LAPS * 500
c.auto_recover = True
#################
trained_pop = []
trained_pop.append(NetworkDriver.from_directory(trained_files.get_file("Human")))
trained_pop.append(NetworkDriver.from_directory(trained_files.get_file("SimpleDriver")))
trained_pop.append(NetworkDriver.from_directory(trained_files.get_file("SimpleExample2")))
trained_pop.append(NetworkDriver.from_directory(trained_files.get_file("bot1")))
trained_pop.append(NetworkDriver.from_directory(trained_files.get_file("bot2")))
evo = Evolution(torcs_client=c)
global_population_fitness_history = []
global_tracks = []
start = timeit.default_timer()
pop = list(trained_pop)
while len(pop) < POPULATION_SIZE:
new_driver = random.choice(trained_pop).mutate()
pop.append(new_driver)
evo.laps = LAPS
evo.track = TRACK
# print "distance raced:", s['distRaced']
return driver.compute_fitness(last_sensor=s, lap_times=lap_times, max_speed=max_speed,
average_speed=avg_speed.avg, timeout_reached=timeout_reached)
else:
return 0.0
except KeyboardInterrupt:
print "Exit client"
except Exception as e:
print "Client Error:", e
finally:
#print "race call took %0.1f seconds." % (timeit.default_timer() - start_time)
self.close()
if __name__ == '__main__':
if len(sys.argv) < 2:
print "Usage: python client.py torcs_install_path"
sys.exit(1)
ndr = NetworkDriver.from_directory(trained_files.get_file("mlp_single_layer"))
# ndr = NetworkDriver.from_file(trained_files.get_file("best_evo_driver_01.pkl"))
#c = Client(torcs_path=sys.argv[1])
#c.configure(track_name='g-track-2', laps=2)
c = Client(torcs_path=None) # Use this for GUI test
c.timeout = 1200
c.auto_recover = False
c.race(driver=ndr)
print "fitness:", ndr.fitness