class Controller:
def __init__(self, problem):
self.load_params()
self._swarm = Swarm(self._num_particles, problem)
def iteration(self, problem):
best_particle = self._swarm.get_best_particle(problem)
for particle in self._swarm:
# best_particle = self._swarm.get_best_neighbour(particle, problem)
particle.update(best_particle, self._w, self._c1, self._c2)
particle.evaluate(problem)
def run(self, problem):
for i in range(self._num_iterations):
self.iteration(problem)
def best_particle(self, problem):
return self._swarm.get_best_particle(problem)
def load_params(self):
f = open("params.in")
self._num_particles = int(f.readline())
self._num_iterations = int(f.readline())
self._w = float(f.readline())
self._c1 = float(f.readline())
self._c2 = float(f.readline())
def __init__(self):
super(MainWindow, self).__init__()
self.swarm = Swarm()
self.ui = Ui_MainWindow()
self.ui.setupUi(self)
self.ui.agregar_final_pushButton.clicked.connect(self.click_agregar)
self.ui.agregar_inicio_pushButton.clicked.connect(
self.click_agregar_inicio)
self.ui.mostrar_pushButton.clicked.connect(self.click_mostrar)
self.ui.actionAbrir.triggered.connect(self.action_abrir_archivo)
self.ui.actionGuardar.triggered.connect(self.action_guardar_archivo)
self.ui.ver_grafos_action.triggered.connect(self.grafos)
self.ui.actionBusqueda.triggered.connect(self.busqueda_grafos)
self.ui.actionId.triggered.connect(self.action_ordenar_id)
self.ui.actionDistancia.triggered.connect(
self.action_ordenar_distancia)
self.ui.actionVelocidad.triggered.connect(
self.action_ordenar_velocidad)
self.ui.mostrar_tabla_pushButton.clicked.connect(self.mostrar_tabla)
self.ui.buscar_pushButton.clicked.connect(self.buscar_iid)
self.ui.dibujar_pushButton.clicked.connect(self.dibujar)
self.ui.limpiar_pushButton.clicked.connect(self.limpiar)
self.scene = QGraphicsScene()
self.ui.graphicsView.setScene(self.scene)
def fitness(particle_position):
"""
Use particle position as parameters for PSO settings, return similarity of
resultant model to Rohde et al. experimental data. Expects 6-dimensional
search space.
"""
global fit_count
# Constants:
dimensions = 3
group_size = 2
no_groups = 100
iterations, particle_settings = get_parameters(particle_position)
particle_settings['respect_boundaries'] = True
scores = []
# run psos with these params
for i, experiment in enumerate(ROHDE_EXPERIMENTS): # TODO: parallelize
print ' ', fit_count, i, '\b' * 100,
swarm = Swarm(dimensions, group_size, no_groups, **particle_settings)
for groups in swarm.step_until(experiment.game, max_iterations=iterations, return_groups=True):
pass
scores.append(experiment.difference(get_coordination_results(groups)))
fit_count += 1
return -1 * sum(scores) / float(len(scores))
def load_data(self, filename):
# do a read from file
self.c1 = 1
self.c2 = 2
self.population_size = 40
self.max_iterations = 40
self.population = Swarm(self.population_size, self.intersection)
def add_peer_to_swarm(self, peer, resource_id, swarm=None):
"""
TODO: implement this method
Based on the resource_id provided, iterate over the
swarms list, and when resource_id matchs, add the
new peer to the swarm.
:param peer: (peer_ip, peer_port)
:param resource_id:
:return: VOID
"""
print(f"Adding peer to {resource_id} swarm")
swarm = swarm or self._get_swarm_object(resource_id)
if not swarm:
swarm = Swarm(resource_id)
self.add_swarm(swarm)
# avoid dups
if not self.peer_in_swarm(peer, swarm):
if swarm.size() < self.MAX_PEERS:
swarm.add_peer(peer)
else:
# too many peers,
print(f"Too many peers in {swarm.resource_id}")
return None
# self.make_persistent()
return True
def test_get_closest_particle():
positions = ([
[-6,0,0],
[-4,0,0],
[-2,0,0],
[3,0,0],
])
velocities = ([
[3,0,0],
[2,0,0],
[1,0,0],
[-1,0,0],
])
accelerations = ([
[0,0,0],
[0,0,0],
[0,0,0],
[0,0,0],
])
s = Swarm(positions, velocities, accelerations)
s.run(10)
assert s.getRemainingParticleCount() == 1
def swarm(t, x, e, swarm_size=1000, omega=0.5):
swarm = Swarm(e, swarm_size)
omega, fip, fig = 0.5, 0.5, 0.5
start = time.time()
while check_time(start, t):
swarm.update_particles_velocity(e, omega, fip, fig)
def swarm(t, x, e, swarm_size=500, omega=0.5, fip=0.9, fig=0.9):
swarm = Swarm(e, swarm_size)
start = time.time()
while check_time(start,t):
swarm.update_particles_velocity(e, omega, fip, fig)
print(swarm.best_swarm_x, swarm.best_swarm_val)
def run(self, flip, max_chance, bees_number, maxIterations, locIterations):
total_time = 0
for itr in range(1, self.nb_exec + 1):
print("Execution {0}".format(str(itr)))
self.fsd = FsProblem(self.typeOfAlgo, self.df, self.ql)
swarm = Swarm(self.fsd, flip, max_chance, bees_number,
maxIterations, locIterations)
t1 = time.time()
best = swarm.bso(self.typeOfAlgo, flip)
t2 = time.time()
total_time += t2 - t1
print("Time elapsed for execution {0} : {1:.2f} s\n".format(
itr, t2 - t1))
self.worksheet.write(itr, 0, itr)
self.worksheet.write(itr, 1, "{0:.2f}".format(best[0]))
self.worksheet.write(itr, 2, best[1])
self.worksheet.write(itr, 3, "{0:.3f}".format(t2 - t1))
self.worksheet.write(
itr, 4, "{0}".format(
str([j[0] for j in [i for i in swarm.best_features()]])))
self.worksheet.write(itr, 5, len(Solution.solutions))
print(
"Total execution time of {0} executions \nfor dataset \"{1}\" is {2:.2f} s"
.format(self.nb_exec, self.dataset_name, total_time))
self.workbook.close()
def metafitness(position,
size=20,
dims=10,
iters=5,
epochs=100,
task_names=None):
# unpack position vector
(
w,
C,
S,
swapping,
velocities,
decrease_velocity,
#add_particle,
replace_particle,
) = position
task_names = task_names if task_names is not None else [
"rastrigin", "ackley", "sphere"
]
task_errors = []
for itask, task_name in enumerate(task_names):
task = fit.string_to_func[task_name]
task_bounds = fit.bounds[task_name]
task_best_loc = fit.actual_minimum(task_name, dims)
task_best_val = task(task_best_loc)
iterations_errors = []
for iter_ in range(iters):
s = Swarm(size,
dims,
fitness=task_name,
bounds=task_bounds,
w=w,
C=C,
S=S,
swapping=swapping,
velocities=velocities,
decrease_velocity=decrease_velocity,
add_particle=0.0,
replace_particle=replace_particle)
result = s.Run(epochs=epochs)
swarm_best_key = sorted(list(result["global_bests"].keys()))[-1]
swarm_best_loc = result["global_bests"][swarm_best_key]
swarm_best_val = task(swarm_best_loc)
error = (task_best_val - swarm_best_val)**2
iterations_errors.append(error)
avg_iterations_error = sum(iterations_errors) / len(iterations_errors)
task_errors.append(avg_iterations_error)
# now we have a list (per-task) of average errors.
# combine *those* with an average (TODO: maybe some consistent scaling?)
# this is our final "meta"-fitness we are trying to optimize.
avg_task_error = sum(task_errors) / len(task_errors)
return avg_task_error
def swarm(t, x, e, swarm_size=100, omega=0.8, fip=1.4, fig=1.3):
swarm = Swarm(e, swarm_size)
start = time.time()
while check_time(start, t):
swarm.update_particles_velocity(e, omega, fip, fig)
print(swarm.best_swarm_x, swarm.best_swarm_val)
sys.stdout.write(str(swarm.best_swarm_val) + " " + str(swarm.best_swarm_x))
def reset(self):
"""
resets the map and the robots for iterative testing/benchmarking
"""
self.grid = GridGraph(self.grid_size)
self.swarm = Swarm(self.num_robots)
self.swarm.startup_sequence(self.grid.list_of_vertices[0])
def __init__(self, intersection):
self.c1 = 1
self.c2 = 2
self.intersection = ""
self.max_iterations = 0
self.population_size = 0
self.load_data("mock file name")
self.population = Swarm(self.c1, self.c2, self.population_size,
intersection)
def runAlg(self):
count = 0
s = Swarm(40, 5, 5)
while s.bestGlobal > 0.000000001:
count += 1
for particle in s.particles:
particle.calFitness()
s.getBestParticle()
for particle in s.particles:
particle.evaluate(s.bestGlobalX, s.bestGlobalY, s.c1, s.c2)
print("Fount in " + str(count) + " moves")
def test_GUI():
swarm = [[0, 10, 6, 5, 11, 4], [0, 3, 9, 1, 8, 7], [0, 0, 0, 0, 2, 0],
[0, 0, 0, 0, 0, 0]]
swarm = np.array(swarm)
shape = [[0, 0, 1, 1, 1], [0, 0, 1, 0, 0], [0, 0, 1, 1, 1],
[0, 0, 1, 0, 0], [0, 0, 1, 1, 1]]
shape = np.array(shape)
S = Swarm(11, swarm)
S.main_sequence(shape, 100, 100, 50)
print(S.moves_sequence)
return instructions_to_GUI(6, S, S.moves_sequence)
class MainGame:
def __init__(self):
config = pyglet.gl.Config(major_version=2, minor_version=1)
self.window = pyglet.window.Window(config=config, width=WIDTH, height=HEIGHT, resizable=False, vsync=True)
#print(self.window.context.get_info().get_renderer())
#print(self.window.context.get_info().get_vendor())
#print('OpenGL Version {}'.format(self.window.context.get_info().get_version()))
# start the background music
#music = pyglet.resource.media('music.mp3')
#music.play()
# manage keys
self.keys = key.KeyStateHandler()
self.window.push_handlers(self.keys)
# init objects
self.camera = Camera(WIDTH, HEIGHT)
self.player = Player()
self.player.Translate(WIDTH/2 - (BLOCK_SIZE * 10)/2, HEIGHT - (BLOCK_SIZE * 10), 0.0)
#self.enemy = Enemy()
#self.enemy.Translate(WIDTH/2 - (BLOCK_SIZE * 10)/2, HEIGHT/2 - (BLOCK_SIZE * 10), 0.0)
self.swarm = Swarm()
# setup function calls
self.window.on_draw = self.on_draw
self.window.on_mouse_press = self.on_mouse_press
self.window.on_mouse_motion = self.on_mouse_motion
pyglet.clock.schedule_interval(self.update, 1/60.0) # update at 60Hz
def on_draw(self):
glViewport(0, 0, WIDTH, HEIGHT)
glClearColor(0.114, 0.114, 0.114, 1.0) #1d1d1d
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT)
self.player.render()
self.swarm.render()
def update(self, dt):
self.player.update(dt)
self.swarm.update(dt)
def on_mouse_press(self, x, y, button, modifiers):
if button == mouse.LEFT:
self.player.fire()
def on_mouse_motion(self, x, y, dx, dy):
# their x and y is from bottom left
# my x and y is from top left
increment_x = x - self.player.position.x
self.player.translate(increment_x, 0.0, 0.0)
def swarm_find_tracker(self, finfo):
peer = self.peer.closest_to_peer(finfo)
range_from_me = self.peer.get_range(finfo)
range_from_peer = Utils.get_range(peer, finfo)
if range_from_me < range_from_peer:
s = Swarm(self.peer.id)
s.create_tracker(s.files[finfo].fstruct)
else:
packet = {
"datatype": "swarm",
"action": "find_tracker",
"finfo": finfo
}
self.peer.client.SendToPeer(peer, packet)
def swarm_find_tracker_red(self, finfo, source):
peer = self.peer.closest_to_peer(finfo)
range_from_me = self.peer.get_range(finfo)
range_from_peer = Utils.get_range(peer, finfo)
if range_from_me < range_from_peer:
packet = {"datatype": "swarm", "action": "found_tracker"}
self.peer.client.SendToPeer(source, packet)
s = Swarm(self.peer.id)
s.create_tracker(None)
else:
packet = {
"datatype": "swarm",
"action": "find_tracker_red",
"source": source
}
def __init__(self):
config = pyglet.gl.Config(major_version=2, minor_version=1)
self.window = pyglet.window.Window(config=config, width=WIDTH, height=HEIGHT, resizable=False, vsync=True)
#print(self.window.context.get_info().get_renderer())
#print(self.window.context.get_info().get_vendor())
#print('OpenGL Version {}'.format(self.window.context.get_info().get_version()))
# start the background music
#music = pyglet.resource.media('music.mp3')
#music.play()
# manage keys
self.keys = key.KeyStateHandler()
self.window.push_handlers(self.keys)
# init objects
self.camera = Camera(WIDTH, HEIGHT)
self.player = Player()
self.player.Translate(WIDTH/2 - (BLOCK_SIZE * 10)/2, HEIGHT - (BLOCK_SIZE * 10), 0.0)
#self.enemy = Enemy()
#self.enemy.Translate(WIDTH/2 - (BLOCK_SIZE * 10)/2, HEIGHT/2 - (BLOCK_SIZE * 10), 0.0)
self.swarm = Swarm()
# setup function calls
self.window.on_draw = self.on_draw
self.window.on_mouse_press = self.on_mouse_press
self.window.on_mouse_motion = self.on_mouse_motion
pyglet.clock.schedule_interval(self.update, 1/60.0) # update at 60Hz
class Controller:
def __init__(self, intersection):
self.c1 = 1
self.c2 = 2
self.intersection = ""
self.max_iterations = 0
self.population_size = 0
self.load_data("mock file name")
self.population = Swarm(self.c1, self.c2, self.population_size,
intersection)
def load_data(self, filename):
# do a read from file
self.c1 = 1
self.c2 = 2
self.population_size = 40
self.max_iterations = 40
self.population = Swarm(self.population_size, self.intersection)
def iteration(self):
for particle in self.population.particles:
particle.update(self.population.get_best_particle())
particle.evaluate()
def run_alg(self):
for i in range(0, self.max_iterations):
self.iteration()
def swarm(t, x, e, swarm_size=1000):
swarm = Swarm(e, swarm_size)
start = time.time()
while check_time(start,t):
for particle in swarm:
def handle_state_announce(self):
assert self.conn_id
if self.conn_id_expires < time.time():
print timestamp(), self, "Connection ID expired"
self.goto_state_conn_id()
return
if not self.current_swarm in [ x for x in Swarm.list() if not x.closed ]:
self.goto_state_swarm()
return
if time.time() < self.announce_next_attempt:
return
retry_text = ''
if self.announce_retry > 0:
retry_text = ', retry %d' % self.announce_retry
print timestamp(), self, "Sending announce for %s%s" % (self.current_swarm, retry_text)
buf = ''
buf += struct.pack('!QLL', self.conn_id, 1, self.transaction_id)
buf += self.current_swarm.sha.decode('hex')
buf += hashlib.sha1(Peer.my_peerid).digest()
buf += struct.pack('!QQQLLLlH', 0, 0, 0, 1, 0, self.key, -1, self.listen_port)
self.send(buf)
self.announce_next_attempt = time.time() + config.tracker_retry_time * 2**min(self.announce_retry, 4)
self.announce_retry += 1
def main():
swarm = Swarm(81, n, 5, -5, 'mesh')
pso = PSO(swarm, P.Function)
pso.Optmize()
x = swarm.particle.bestPosition
def from_swarm_dicts(cls,
swarm_dicts,
inference_model,
scoring_functions=None,
x_min=-1.,
x_max=1.,
inertia_weight=0.9,
phi1=2.,
phi2=2.,
phi3=2.,
**kwargs):
"""
Classmethod to create a PSO instance from a list of dictionaries each defining an
individual swarm.
:param swarm_dicts: A list of dictionaries each defining an individual swarm.
See Swarm.from_dict for more info.
:param inference_model: A inference model instance that is used for encoding an decoding
SMILES to and from the CDDD space.
:param scoring_functions: List of functions that are used to evaluate a generated molecule.
Either take a RDKit mol object as input or a point in the cddd space.
:param kwargs: additional parameters for the PSO class
:return: A PSOptimizer instance.
"""
swarms = [
Swarm.from_dict(dictionary=swarm_dict,
x_min=x_min,
x_max=x_max,
inertia_weight=inertia_weight,
phi1=phi1,
phi2=phi2,
phi3=phi3) for swarm_dict in swarm_dicts
]
return cls(swarms, inference_model, scoring_functions, **kwargs)
def run_task(t: Task):
try:
strategy, iterations, num_good_clients, num_free_riders, peer_size, max_up, max_down = t
swarm = Swarm()
all_agents = chain(
(Client(
strat=strategy,
up=Points(max_up),
down=Points(max_down),
peer_size=peer_size,
swarm=swarm,
iterations=iterations
) for _ in range(num_good_clients)),
(Client(
strat=strategy,
up=no_points,
down=Points(max_down),
peer_size=peer_size,
swarm=swarm,
iterations=iterations
) for _ in range(num_free_riders))
)
[swarm.join(x) for x in all_agents]
OUTPUT_FILE = f"./results/{strategy.__name__}/{iterations}_{num_good_clients}_{num_free_riders}_{peer_size}_{max_up}_{max_down}.json"
if not os.path.isdir(os.path.dirname(OUTPUT_FILE)):
os.makedirs(os.path.dirname(OUTPUT_FILE))
with open(OUTPUT_FILE, 'w') as f:
f.write(dumps(
{
'metadata': {
'strategy': strategy.__name__,
'iterations': iterations,
'max_up': max_up,
'max_down': max_down,
'starting_good_clients': num_good_clients,
'starting_bad_clients': num_free_riders,
'peer_size': peer_size
},
'data': tuple(
chain.from_iterable(
(x.to_json(iteration) for x in y) for iteration, y in enumerate(Model.run(swarm, iterations))))
}
))
except Exception as e:
print("got exception ", e)
def from_query_list(cls,
init_smiles,
num_part,
num_swarms,
inference_model,
scoring_functions=None,
phi1=2.,
phi2=2.,
phi3=2.,
x_min=-1.,
x_max=1.,
v_min=-0.6,
v_max=0.6,
**kwargs):
"""
Classmethod to create a PSO instance with (possible) multiple swarms which particles are
initialized at the position of the embedded input SMILES. Each swarms is initialized at
the position defined by the different SMILES in the input list.
:param init_smiles: A List of SMILES which each define the molecule which acts as starting
point of each swarm in the optimization.
:param num_part: Number of particles in each swarm.
:param num_swarms: Number of individual swarm to be optimized.
:param inference_model: A inference model instance that is used for encoding an decoding
SMILES to and from the CDDD space.
:param scoring_functions: List of functions that are used to evaluate a generated molecule.
Either take a RDKit mol object as input or a point in the cddd space.
:param phi1: PSO hyperparamter.
:param phi2: PSO hyperparamter.
:param phi3: PSO hyperparamter.
:param x_min: min bound of the optimization space (should be set to -1 as its the default
CDDD embeddings take values between -1 and 1).
:param x_max: max bound of the optimization space (should be set to -1 as its the default
CDDD embeddings take values between -1 and 1).
:param v_min: minimal velocity component of a particle. Also used as lower bound for the
uniform distribution used to sample the initial velocity.
:param v_max: maximal velocity component of a particle. Also used as upper bound for the
uniform distribution used to sample the initial velocity.
:param kwargs: additional parameters for the PSO class
:return: A PSOptimizer instance.
"""
assert isinstance(init_smiles, list)
assert len(init_smiles) == num_swarms
embedding = inference_model.seq_to_emb(init_smiles)
swarms = []
for i, sml in enumerate(init_smiles):
swarms.append(
Swarm.from_query(init_sml=sml,
init_emb=embedding[i],
num_part=num_part,
v_min=v_min,
v_max=v_max,
x_min=x_min,
x_max=x_max,
phi1=phi1,
phi2=phi2,
phi3=phi3))
return cls(swarms, inference_model, scoring_functions, **kwargs)
def __init__(self, function_to_be_minimized, constants):
self.constants = constants
self.precision = constants.PRECISION
self.inertia = constants.INERTIA_WEIGHT
self.cognitive = constants.COGNITIVE_WEIGHT
self.social = constants.SOCIAL_WEIGHT
self.max_velocity_allowed = constants.MAX_VELOCITY_ALLOWED
self.pop_size = constants.POP_SIZE
self.generations_no = constants.GENERATIONS_NO
self.dimensions_no = constants.DIMENSIONS_OF_THE_FUNCTION
self.fitness = function_to_be_minimized
self.a = constants.INTERVALS_OF_DEFINITION[self.fitness.__name__][0]
self.b = constants.INTERVALS_OF_DEFINITION[self.fitness.__name__][1]
self.population = []
self.swarm = Swarm(self.fitness, self.precision)
def on_heartbeat(self):
# Kludge for now. Just have every tracker announce every swarm
for x in Swarm.list():
self.add_swarm(x)
# Cleanup old swarms
for x in [x for
x in self.swarms.keys()
if x.closed or x not in Swarm.list()]:
print timestamp(), self, "Swarm seems to have gone away, removing:", x
del self.swarms[x]
if self.state == STATE_SOCK: self.handle_state_sock()
elif self.state == STATE_SWARM: self.handle_state_swarm()
elif self.state == STATE_CONN_ID: self.handle_state_conn_id()
elif self.state == STATE_ANNOUNCE: self.handle_state_announce()
elif self.state == STATE_MUTE: self.handle_state_mute()
else: raise Exception("State machine broken?")
def main():
options, files = getopt.getopt(sys.argv[1:], 'ivt:', ['info', 'verbose', 'threads:'])
for flag, value in options:
if flag == '-v' or flag == '--verbose':
configuration['verbose'] = True
elif flag == '-t' or flag == '--threads':
try:
configuration['threads'] = int(value)
except:
usage()
elif flag == '-i' or flag == '--info':
configuration['info'] = True
else:
usage()
if len(files) != 1:
usage()
try:
torrent = Torrent(files[0])
except:
print 'Impossible to read torrent file/magnet link:', files[0]
exit(1)
if configuration['info'] == True:
torrent.show()
exit(0)
torrent.start()
generate_clientid()
tracker = Tracker(torrent.data['announce'])
tracker.update(torrent)
swarm = Swarm()
swarm.update_peers(tracker)
threads = configuration['threads']
def test_get_closest_particle():
positions = ([
[3, 0, 0],
[4, 0, 0],
])
velocities = ([
[2, 0, 0],
[0, 0, 0],
])
accelerations = ([
[-1, 0, 0],
[-2, 0, 0],
])
s = Swarm(positions, velocities, accelerations)
s.run(10)
assert s.getClosestParticleIndex() == 0
def main():
positions = []
velocities = []
accelerations = []
line = f.readline()
while line:
m = re.search('p=<(.*)>, v=<(.*)>, a=<(.*)>', line.rstrip('\n'))
pStrings = m.group(1).split(',')
vStrings = m.group(2).split(',')
aStrings = m.group(3).split(',')
positions.append([int(pStrings[0]), int(pStrings[1]), int(pStrings[2])])
velocities.append([int(vStrings[0]), int(vStrings[1]), int(vStrings[2])])
accelerations.append([int(aStrings[0]), int(aStrings[1]), int(aStrings[2])])
line = f.readline()
s = Swarm(positions, velocities, accelerations)
s.run(1000)
print(s.getRemainingParticleCount())
def pso(self, w=0.45, c1=1.70, c2=2.49):
time_to_end = time.time() + self.t
# C₁ importance of personal best value
# C₂ importance of social best value.
# w meaning of velocity vector
swarm = Swarm(self.SWARM_SIZE, self.VECTOR_SIZE, self.params, w, c1,
c2)
while time.time() < time_to_end:
swarm.update_swarm()
swarm.set_best_particle()
#print(swarm.best_particle)
#print("#BEST", swarm.best_particle)
return swarm.best_particle
def main():
metaswarm = Swarm(size=25,
dims=len(meta_params),
fitness="sphere",
bounds=(0.0, 1.0))
metaswarm.fitnessFunc = metafitness
metaswarm._fitnessString = "metafitness"
metaswarm.CheckConvergence = _dummy_false
results = metaswarm.Run(100)
results.save_json("metaswarm_output.json")
best_settings_key = sorted(list(results["global_bests"].keys()))[-1]
best_settings = results["global_bests"][best_settings_key]
for setting, value in zip(meta_params, best_settings):
print(f"{setting}\t{value}")
class Main(object):
"""
the Main running class, mainly used for running the algorithm for accuracy
and for benchmarking its efficiency. visualization.py was used for actual visualization
"""
def __init__(self, grid_size, num_robots):
self.grid_size, self.num_robots = grid_size, num_robots
self.grid = GridGraph(self.grid_size)
self.swarm = Swarm(self.num_robots)
self.swarm.startup_sequence(self.grid.list_of_vertices[0])
def reset(self):
"""
resets the map and the robots for iterative testing/benchmarking
"""
self.grid = GridGraph(self.grid_size)
self.swarm = Swarm(self.num_robots)
self.swarm.startup_sequence(self.grid.list_of_vertices[0])
def run(self):
"""
runs the algorithms until complete, and keeps track of the number of times it calls update,
which is considered our benchmark
"""
profile = 0
while not (all([
True if robot.state == "standby" else False
for robot in self.swarm.swarm
]) and (not [
area
for area in self.swarm.unknown_territory if area.state == "red"
])):
self.swarm.update()
profile += 1
# makes sure that the map we made is correct to the environment's map
actual_graph = {
vertex.name: {neighbor.name
for neighbor in vertex.neighbors}
for vertex in self.grid.list_of_vertices
}
if actual_graph == self.swarm.map:
return profile
else:
raise ValueError("You done g00fed")
def startSwarm(self, size = None):
'''
Runs permutations on model parameters to find the optimal model
characteristics for the given data.
* size - A value from grokpy.SwarmSize. Initially, small, medium, or large.
the default is medium. Small is only good for testing, whereas
large can take a very long time.
'''
if self.swarm and self.swarm.getState() in ['Starting', 'Running']:
raise GrokError('This model is already swarming.')
url = self.swarmsUrl
requestDef = {}
if size:
requestDef.update({'options': {"size": size}})
result = self.c.request('POST', url, requestDef)
# Save the swarm object
self.swarm = Swarm(self, result['swarm'])
return result
def __init__(self):
#Start Pygame
pygame.init()
pygame.display.set_caption('Graph Exploration')
#Create Pygame variables
self.clock = pygame.time.Clock()
self.sizex = 700
self.sizey = 700
self.imgCounter = 0
self.font = pygame.font.SysFont('Roboto', 25)
self.background_color = (227, 232, 239)
self.robot_color = (83, 87, 94)
self.edge_color = (83, 87, 94)
self.state_to_color_mapping = {
'red': pygame.Color('red'),
'yellow': pygame.Color('yellow'),
'green': pygame.Color('green')
}
self.screen = pygame.display.set_mode((self.sizex, self.sizey))
self.background = pygame.Surface((self.sizex, self.sizey))
#Create Grid for robots to walk on
self.grid = TripleGraph()
#Create swarm of robots
self.swarm = Swarm(2)
self.swarm.startup_sequence(self.grid.list_of_vertices[0])
self.old_vertices = None
swarm = Swarm(3)
swarm.startup_sequence(self.grid.list_of_vertices[0])
#Set Main loop to running
self._running = True
#Draw graph into window
self.background.fill(self.background_color)
self.space_out_vertices(self.grid)
self.draw_grid()
pygame.draw.circle(self.background, self.robot_color,
self.swarm.hive.coords, 15, 3)
self.screen.blit(self.background, (0, 0))
pygame.display.update()
sleep(1)
def run(swarm: Swarm, iterations: int) -> Iterator[Iterator[Result]]:
all_agents = list(swarm.all_clients())
[x.init_peers() for x in all_agents]
for c in range(iterations):
print("Iteration", c)
do_assertions(all_agents)
remaining_agents = set(all_agents)
while remaining_agents:
# filter out the ones that don't want content
remaining_agents -= {x for x in all_agents if not x.wants_content()}
# Iterate in random order
for agent in random_iteration(remaining_agents):
# Iterate all peers of that agent in random order
for peer in random_iteration(agent.peers):
if peer.ask_for_content(agent): # If they gave us content
agent.give_content(peer)
break
else:
# None of the peers gave them something
remaining_agents.remove(agent)
yield (x.get_state() for x in all_agents)
[x.reset_values() for x in all_agents]
do_assertions(all_agents)
# Find new peers
print("--before--")
[x.before_reset() for x in random_iteration(all_agents)]
print("--reset--")
[x.reset(c) for x in random_iteration(all_agents)]
print("--after--")
[x.after_reset(c) for x in random_iteration(all_agents)]
do_assertions(all_agents)
"respect_boundaries"
] = True # Actually seems to track more closely without this, just at much lower (1/10th) rates
# particle_settings['initial_inertia'] = 0.00019464380975982382 # Brennan & Clark 1996
graph = False
if graph:
from matplotlib import pyplot as pl
from matplotlib import animation
from mpl_toolkits.mplot3d import Axes3D
for i, experiment in enumerate(parameter_estimation.ROHDE_EXPERIMENTS):
fig = pl.figure()
ax = fig.add_subplot(111, projection="3d")
# pl.axis([-0.05, 1.05,] * dimensions)
swarm = Swarm(dimensions, group_size, no_groups, **particle_settings)
# plot = pl.scatter(*zip(*[particle.position for particle in swarm.step(experiment.game)]), alpha=0.2)
plot = pl.scatter(*zip(*[particle.position for particle in swarm.step(experiment.game)]))
anim = animation.FuncAnimation(
fig, update_plot, frames=xrange(iterations), fargs=(swarm, plot, experiment.game)
)
pl.show()
else:
# new_experiments = [
# Experiment( # free ambiguous form
# Experiment.Settings(
# reference_costs=(-60., -120., -280.),
# ambiguous_reference_cost=-0.,
if __name__ == '__main__':
output_location = r'I:\Users\Chase Stevens\Dropbox\Dissertation\swarm_state.json'
iterations = 10000
dimensions = 6
group_size = 100
no_groups = 1
save = True
graph = False
if os.path.exists(output_location):
with open(output_location, 'r') as f:
swarm = Swarm.from_dict(json.load(f))
print "Loaded swarm - resuming from iteration", swarm.particles[0]._time
print "(Loaded", len(swarm.particles), "particles into", len(swarm.particle_groups), "groups)"
else:
swarm = Swarm(
dimensions,
group_size,
no_groups,
particle_distribution=parameter_estimation.mitchell_sampling_factory(dimensions),
)
fitness_func = parameter_estimation.fitness
if graph:
from matplotlib import pyplot as pl
from matplotlib import animation
class Model(object):
'''
Object representing a Grok Model.
* parent - **Either** a `Client` or `Project`.
* modelDef - A dict, usually returned from a model creation or get action.
Usually includes:
* id
* name
* streamId
* swarmsUrl
* url
'''
def __init__(self, parent, modelDef):
# Give streams access to the parent client/project and its connection
self.parent = parent
self.c = self.parent.c
# Take everything we're passed and make it an instance property.
self.__dict__.update(modelDef)
# Prepare to have a swarm associated with the model
self.swarm = None
def _runCommand(self, command, **params):
url = self.commandsUrl
commandObject = {'command': command}
if params:
commandObject['params'] = params
result = self.c.request('POST', url, commandObject)
return result
def delete(self):
'''
Permanently deletes the model
.. warning:: There is currently no way to recover from this opperation.
'''
self.c.request('DELETE', self.url)
def clone(self, params = None):
'''
Clones this model
'''
if params:
result = self.c.request('POST', self.cloneUrl, {'model': params})
else:
result = self.c.request('POST', self.cloneUrl)
return Model(self.parent, result['model'])
#############################################################################
# Model Configuration
def setName(self, newName):
'''
Renames the model.
* newName - String
'''
# Get the current model state
modelDef = self.getState()
# Update the definition
modelDef['name'] = newName
# Update remote state
self.c.request('POST', self.url, {'model': modelDef})
# Update local state
self.name = newName
def setNote(self, newNote):
'''
Adds or updates a note for this model.
* newNote - A String describing this model.
'''
# Get the current model state
modelDef = self.getState()
# Update the definition
modelDef['note'] = newNote
# Update remote state
self.c.request('POST', self.url, {'model': modelDef})
# Update local state
self.note = newNote
#############################################################################
# Model states
def getState(self):
'''
Returns the full state of the model from the API server
TODO: Remove popping of Nones once API is updated
'''
modelDef = self.c.request('GET', self.url)['model']
# API doesn't except None values. Remove them.
for key, value in modelDef.items():
if value is None:
del modelDef[key]
return modelDef
def promote(self, **params):
'''
Puts the model into a production ready mode.
.. note: This may take several seconds.
'''
# Check how many records are in the swarm model output cache
headers, data, meta = self.getModelOutput()
swarmOutputCacheLen = len(data)
# Promotion command
self._runCommand('promote', **params)
##### WARNING: Ugly hack that will go away #####
# Check if the state of the model changes
timeoutCounter = 0
while True:
modelDef = self.c.request('GET', self.url)['model']
status = modelDef['status']
if timeoutCounter >= 20:
raise GrokError('The production model did not start in a reasonable '
'amount of time. Please try again or contact support.')
elif status == 'running':
# The production model has turned on
break
else:
print 'Waiting for the production model to become ready ...'
time.sleep(.5)
timeoutCounter += 1
# Check that the production model has at least caught up to where we were
# at the end of swarm.
timeoutCounter = 0
while True:
# Production Model Output Cache Length
try:
headers, data, meta = self.getModelOutput()
pmocLen = len(data)
except requests.exceptions.HTTPError:
print 'Whoops 500'
time.sleep(.5)
timeoutCounter += 1
continue
if timeoutCounter >= 80:
raise GrokError("The production model did not catch up to the swarm "
"in a reasonable amount of time. Please try again or "
"contact support.")
elif pmocLen >= swarmOutputCacheLen:
break
else:
print 'Waiting for the production model to catch up with the data ...'
time.sleep(1)
timeoutCounter += 1
##### END UGLY HACK
def start(self, **params):
'''
Starts up a model, readying it to receive new data from a stream
'''
return self._runCommand('start', **params)
def disableLearning(self, **params):
'''
Puts the model into a predictions only state where it will not learn from
new data.
.. note:: This method is intended for use with RUNNING models that have
been promoted. The API server will return an error in other cases.
'''
return self._runCommand('disableLearning', **params)
def enableLearning(self, **params):
'''
New records will be integrated into the models future predictions.
.. note:: This method is intended for use with RUNNING models that have
been promoted. The API server will return an error in other cases.
'''
return self._runCommand('enableLearning', **params)
def setAnomalyAutoDetectThreshold(self, autoDetectThreshold):
'''
Sets the autoDetectThreshold of the model. Model must be TemporalAnomaly
for this to succeed.
* autoDetectThreshold - value to set the auto detect threshold
.. note:: This method is intended for use with RUNNING models that have
been promoted. The API server will return an error in other cases.
'''
return self._runCommand('setAutoDetectThreshold', \
autoDetectThreshold=autoDetectThreshold)
def getAnomalyAutoDetectWaitRecords(self):
'''
Gets the autoDetectWaitRecords of the model. Model must be TemporalAnomaly
for this to succeed.
Response on success::
{
'autoDetectWaitRecords': integer
}
.. note:: This method is intended for use with RUNNING models that have
been promoted. The API server will return an error in other cases.
'''
return self._runCommand('getAutoDetectWaitRecords')
def setAnomalyAutoDetectWaitRecords(self, autoDetectWaitRecords):
'''
Sets the autoDetectWaitRecords of the model. Model must be TemporalAnomaly
for this to succeed.
* autoDetectWaitRecords - value to set the auto detect wait records
.. note:: This method is intended for use with RUNNING models that have
been promoted. The API server will return an error in other cases.
'''
return self._runCommand('setAutoDetectWaitRecords', \
autoDetectWaitRecords=autoDetectWaitRecords)
def getAnomalyAutoDetectThreshold(self):
'''
Gets the autoDetectThreshold of the model. Model must be TemporalAnomaly
for this to succeed.
Response on success::
{
'autoDetectThreshold': float
}
.. note:: This method is intended for use with RUNNING models that have
been promoted. The API server will return an error in other cases.
'''
return self._runCommand('getAutoDetectThreshold')
def getLabels(self, startRecordID=None, endRecordID=None):
'''
Returns a list of labels for a given range of records. Each record has a
list of labels assocciated. A label may have no labels.
* startRecordID - ROWID of the first prediction of these label results
* endRecordID - ROWID of the last prediction record of these label results.
(Not inclusive.)
Response on success::
{
'isProcessing': boolean,
'recordLabels': [
{
'ROWID': integer,
'labels': [str, ...]
},...
]
}
.. note:: This method is intended for use with RUNNING models that have
been promoted. The API server will return an error in other cases.
'''
return self._runCommand('getLabels', startRecordID=startRecordID,
endRecordID=endRecordID)
def addLabel(self, startRecordID, endRecordID, labelName):
'''
Adds a label to a given range of records from startRecordID to endRecordID,
not inclusive of endRecordID.
* startRecordID - ROWID of the first prediction to add this label
* endRecordID - ROWID of the last prediction record to add this label.
(Not inclusive.)
* labelName - string indicating name of the label to add to the given range
Response on success::
{
'status': 'success'
}
.. note:: This method is intended for use with RUNNING models that have
been promoted. The API server will return an error in other cases.
'''
return self._runCommand('addLabel', startRecordID=startRecordID, \
endRecordID=endRecordID, labelName=labelName)
def removeLabels(self, startRecordID=None, endRecordID=None, labelFilter=None):
'''
Removes a label or all labels from a given range of records from
startRecordID to endRecordID, not inclusive of endRecordID. If labelFilter
is set, only labels of type labelFilter will be removed, otherwise all
labels will be removed from the given range.
* startRecordID - ROWID of the first prediction to remove labels
* endRecordID - ROWID of the last prediction record to remove labels.
(Not inclusive.)
* labelFilter - string. If not None, only labels equal to this will be
removed. Otherwise all labels will be removed from given
range.
Response on success::
{
'status': 'success'
}
.. note:: This method is intended for use with RUNNING models that have
been promoted. The API server will return an error in other cases.
'''
return self._runCommand('removeLabels', startRecordID=startRecordID, \
endRecordID=endRecordID, labelFilter=labelFilter)
#############################################################################
# Stream
def getStream(self):
'''
Returns the Stream that this model is associated with.
'''
return self.parent.getStream(self.streamId)
#############################################################################
# Swarms
def listSwarms(self):
'''
Returns a list of Swarm objects for this model
'''
# Where to make our request
url = self.swarmsUrl
swarmDefs = self.c.request('GET', url)['swarms']
swarms = []
for swarmDef in swarmDefs:
swarms.append(Swarm(self, swarmDef))
return swarms
def startSwarm(self, size = None):
'''
Runs permutations on model parameters to find the optimal model
characteristics for the given data.
* size - A value from grokpy.SwarmSize. Initially, small, medium, or large.
the default is medium. Small is only good for testing, whereas
large can take a very long time.
'''
if self.swarm and self.swarm.getState() in ['Starting', 'Running']:
raise GrokError('This model is already swarming.')
url = self.swarmsUrl
requestDef = {}
if size:
requestDef.update({'options': {"size": size}})
result = self.c.request('POST', url, requestDef)
# Save the swarm object
self.swarm = Swarm(self, result['swarm'])
return result
def stop(self):
'''
Stops (and checkpoints) a running model. If a swarm is in progress it will
gracefully stop the swarm progress and make the model available for
promotion (this can take a few seconds).
'''
self._runCommand('stop')
def getSwarmState(self):
'''
Returns the current state of a swarm.
'''
if not self.swarm:
# See if the API knows about a swarm for this model
swarms = self.listSwarms()
if not swarms:
raise GrokError('There is no swarm associated with this model.')
else:
# Find the latest swarm
# Use the string time to sort with the latest time first
# If details hasn't been populated yet, assume this is the latest
swarms = sorted(swarms,
key = lambda swarm: swarm.details.get('startTime', 0),
reverse=True)
# Set that as *the* swarm for this model
self.swarm = swarms[0]
return self.swarm.getState()
#############################################################################
# Checkpoints
def listCheckpoints(self):
'''
Returns a list of checkpoint dicts for this model
'''
# Where to make our request
url = self.checkpointsUrl
checkpoints = self.c.request('GET', url)['checkpoints']
return checkpoints
def createCheckpoint(self):
'''
Creates a new checkpoint object and returns it as a dict
'''
# Where to make our request
url = self.checkpointsUrl
checkpoint = self.c.request('POST', url)['checkpoint']
return checkpoint
#############################################################################
# Model data
def getModelOutput(self, limit=None, offset = None,
startAt = None, shift = True):
'''
Returns the data in the output cache of the best model found during Swarm.
* limit - The maximum number of rows to get from the model
* offset - The number of rows from the last row from which to begin
returning data from.
For example::
If you have 1000 records in the model output cache and set offset to
100, you will get records with row ID 900 to 999. If you set offset
above the maximum row ID that exists in the model's output cache you will
get no records.
* startAt - The start row ID to begin returning data from.
For example::
If you have 1000 records in the model output cache and set startAt to
100, you will get records with row ID 100 to 1000. If you set startAt
above the maximum row ID that exists in the model's output cache you will
get no records.
* shift - This shifts the records returned so that all predictions are
aligned with actual values. Note: Set this value to False if you
are working with realtime data.
'''
params = {}
if limit is not None:
params['limit'] = limit
if offset is not None:
params['offset'] = offset
if startAt is not None:
params['startAt'] = startAt
if shift is not None:
params['shift'] = shift
result = self.c.request('GET', self.dataUrl, params = params)['output']
headers = result['names']
data = result['data']
try:
meta = result['meta']
except KeyError:
meta = None
return headers, data, meta
scale = numpy.array([5, 10])
plot.set_offsets(zip([particle.position * scale for particle in swarm.step(fitness_func)]))
return plot,
if __name__ == '__main__':
iterations = 600
dimensions = 2
group_size = 50
no_groups = 50
save = True
graph = True
swarm = Swarm(dimensions, group_size, no_groups, respect_boundaries=False, velocity_dampening=0.2)
fitness_func = lambda (x, y): -abs((y * 10) - (x * 5) ** 2)
if graph:
from matplotlib import pyplot as pl
from matplotlib import animation
fig = pl.figure()
#ax = pl.axis([0, 10,] * dimensions)
ax = pl.axis([0, 5, 0, 10])
plot = pl.scatter(*zip(*[particle.position for particle in swarm.step(fitness_func)]))
anim = animation.FuncAnimation(fig, update_plot, frames=iterations, fargs=(swarm, plot, fitness_func))
anim.save('squares.mp4', fps=10, extra_args=['-vcodec', 'libx264'])
def main():
# Setup some basic logging
logging.basicConfig()
formatter = logging.Formatter('%(name)-24s: %(asctime)s : %(levelname)-8s %(message)s')
logger = logging.getLogger()
logger.handlers[0].setFormatter(formatter)
# Callback for showing statistics
def log_stats(data):
sideband = 'USB'
auto_amps = {}
for baseline in data.baselines:
if baseline.is_valid():
chunk = baseline.left._chk
interleaved = array(list(p for p in data[baseline][chunk][sideband] if not isnan(p)))
complex_data = interleaved[0::2] + 1j * interleaved[1::2]
if baseline.is_auto():
auto_amps[baseline] = abs(complex_data).mean()
norm = auto_amps[baseline]
else:
norm_left = auto_amps[SwarmBaseline(baseline.left, baseline.left)]
norm_right = auto_amps[SwarmBaseline(baseline.right, baseline.right)]
norm = sqrt(norm_left * norm_right)
logger.info(
'{baseline!s}[chunk={chunk}].{sideband} : Amp={amp:.4e}, Phase={pha:+.2e}, Corr.={corr:5.2f}%'.format(
baseline=baseline, chunk=chunk, sideband=sideband,
corr=100.0*abs(complex_data).mean()/norm,
pha=angle(complex_data).mean(),
amp=abs(complex_data).mean()
)
)
# Callback for saving raw data
def save_rawdata(rawdata):
for fid, datas in enumerate(rawdata):
filename = 'fid%d.dat' % fid
save_bin(filename, datas)
logger.info('Data for FID #%d saved to %r' % (fid, filename))
# Callback for checking ramp
def check_ramp(rawdata):
ramp = empty(SWARM_VISIBS_ACC_SIZE)
for fid, datas in enumerate(rawdata):
raw = array(unpack('>%dI'%SWARM_VISIBS_ACC_SIZE, datas))
ramp[0::2] = raw[1::2]
ramp[1::2] = raw[0::2]
errors_ind = list(i for i, p in enumerate(ramp) if p !=i)
logger.info('Ramp errors for FID #%d: %d' % (fid, len(errors_ind)))
# Parse the user's command line arguments
parser = argparse.ArgumentParser(description='Catch and process visibility data from a set of SWARM ROACH2s')
parser.add_argument('-v', dest='verbose', action='store_true', help='display debugging logs')
parser.add_argument('-m', '--swarm-mapping', dest='swarm_mapping', metavar='SWARM_MAPPING', type=str, default=SWARM_MAPPING,
help='Use file SWARM_MAPPING to determine the SWARM input to IF mapping (default="%s")' % SWARM_MAPPING)
parser.add_argument('-i', '--interface', dest='interface', metavar='INTERFACE', type=str, default='eth2',
help='listen for UDP data on INTERFACE (default="eth2")')
parser.add_argument('-b', '--bitcode', dest='bitcode', metavar='BITCODE', type=str, default=DEFAULT_BITCODE,
help='program ROACH2s with BITCODE (default="%s")' % DEFAULT_BITCODE)
parser.add_argument('-t', '--integrate-for', dest='itime', metavar='INTEGRATION-TIME', type=float, default=30.0,
help='integrate for approximately INTEGRATION-TIME seconds (default=30)')
parser.add_argument('--setup-only', dest='setup_only', action='store_true',
help='only program and setup the board; do not wait for data')
parser.add_argument('--listen-only', dest='listen_only', action='store_true',
help='do NOT setup the board; only wait for data')
parser.add_argument('--visibs-test', dest='visibs_test', action='store_true',
help='enable the DDR3 visibility ramp test')
parser.add_argument('--save-raw-data', dest='save_rawdata', action='store_true',
help='Save raw data from each FID to file')
parser.add_argument('--log-stats', dest='log_stats', action='store_true',
help='Print out some baselines statistics (NOTE: very slow!)')
args = parser.parse_args()
# Set logging level given verbosity
if args.verbose:
logger.setLevel(logging.DEBUG)
else:
logger.setLevel(logging.INFO)
# Silence katcp INFO messages
katcp_logger = logging.getLogger('katcp')
katcp_logger.setLevel(logging.WARNING)
# Setup the listener class
listener = SwarmListener(args.interface)
# Create our SWARM instance
swarm = Swarm()
if not args.listen_only:
# Setup using the Swarm class and our parameters
swarm.setup(args.bitcode, args.itime, listener)
if args.visibs_test:
# Enable the visibs test
swarm.members_do(lambda fid, member: member.visibs_delay(delay_test=True))
else:
# Disable the visibs test
swarm.members_do(lambda fid, member: member.visibs_delay(delay_test=False))
if not args.setup_only:
# Create the data handler
swarm_handler = SwarmDataHandler(swarm, listener)
if args.log_stats:
# Use a callback to show visibility stats
swarm_handler.add_callback(log_stats)
if args.save_rawdata:
# Give a rawback that saves the raw data
swarm_handler.add_rawback(save_rawdata)
if args.visibs_test:
# Give a rawback that checks for ramp errors
swarm_handler.add_rawback(check_ramp)
# Start the main loop
swarm_handler.loop()