class VehicleLaneAlignmentNetworkIn(BaseNetwork):
def __init__(self, topic='/laneletInformation'):
super(VehicleLaneAlignmentNetworkIn, self).__init__()
self._vehicle_alignment_pop = Population(1, IF_curr_alpha, {'i_offset': 0, 'v_thresh': 100,
'tau_refrac': 0.1})
detector = nest.nest.Create('spike_detector', params={'withgid': True, 'withtime': True})[0]
self.detectors[self._vehicle_alignment_pop[0]] = detector
nest.nest.Connect(self._vehicle_alignment_pop[0], detector)
self._state = None
self._sub_lanelet = rospy.Subscriber(topic, Float64MultiArray, self._update_state, queue_size=1)
self.output_pop = self._vehicle_alignment_pop
# make sure we get at least one message before starting the training
rospy.wait_for_message(topic, Float64MultiArray)
def _update_state(self, state):
if len(state.data) > 0:
self._state = LaneletInformation(state.data)
def before_simulation(self):
i_offset = np.pi - np.abs(self._state.angle_vehicle_lane)
i_offset *= 30/np.pi
print "I-Offset Vehicle Alignment: ", i_offset
self._vehicle_alignment_pop.set(i_offset=i_offset)
def main(args):
setup(timestep=0.1)
random_image = np.random.rand(2,2)
size = random_image.size
input_population_arr = Population(random_image.size, SpikeSourceArray, {'spike_times': [0 for i in range(0, random_image.size)]})
cell_params = {'tau_refrac': 2.0, 'v_thresh': -50.0, 'tau_syn_E': 2.0, 'tau_syn_I': 2.0}
output_population = Population(1, IF_curr_alpha, cell_params, label="output")
projection = Projection(input_population_arr, output_population, AllToAllConnector())
projection.setWeights(1.0)
input_population_arr.record('spikes')
output_population.record('spikes')
tstop = 1000.0
run(tstop)
output_population.write_data("simpleNetwork_output.pkl",'spikes')
input_population_arr.write_data("simpleNetwork_input.pkl",'spikes')
#output_population.print_v("simpleNetwork.v")
end()
def createNetwork():
# the dimesion is X x Y x maxDisparity+1 because disparity 0 means no shift in pixel location
# however the network should still exist and contain only one disparity map.
assert dimensionRetinaX > maxDisparity >= 0, "Maximum Disparity Constant is illegal!"
print "Creating Cooperative Network..."
print "\t Creating Populations..."
from SimulationAndNetworkSettings import t_synE, t_synI, t_memb, vResetInh, vResetCO
from pyNN.nest import record
network = []
for x in range(0, dimensionRetinaX):
for disp in range(0, maxDisparity+1):
inhLeftPop = Population(dimensionRetinaY, IF_curr_exp, {'tau_syn_E':t_synE, 'tau_syn_I':t_synI, 'tau_m':t_memb, 'v_reset':vResetInh},
label="Blocker Left {0}".format(x*dimensionRetinaX + disp))
inhRightPop = Population(dimensionRetinaY, IF_curr_exp, {'tau_syn_E':t_synE, 'tau_syn_I':t_synI, 'tau_m':t_memb, 'v_reset':vResetInh},
label="Blocker Right {0}".format(x*dimensionRetinaX + disp))
cellOutputPop = Population(dimensionRetinaY, IF_curr_exp, {'tau_syn_E':t_synE, 'tau_syn_I':t_synI, 'tau_m':t_memb, 'v_reset':vResetCO},
label="Cell Output {0}".format(x*dimensionRetinaX + disp))
# reocrd data for plotting purposes
cellOutputPop.record('v')
network.append((inhLeftPop, inhRightPop, cellOutputPop))
interconnectNetworkNeurons(network)
return network
def test_set_with_standard_celltype(self):
p = Population(10, MockStandardCellType)
p.all_cells = numpy.array([MockID()]*10, dtype=object) #numpy.arange(10)
p._mask_local = numpy.ones((10,), bool)
p.set("foo", 32)
assert_equal(nest.SetStatus.call_args[0][1], {"FOO": 32.0})
p.set("hoo", 33.0)
assert_equal(nest.SetStatus.call_args[0][1], {"HOO": 99.0})
p.set("woo", 6.0)
assert_equal(nest.SetStatus.call_args[0][1], {})
p.all_cells[0].set_parameters.assert_called_with(woo=6.0)
def test_set_with_native_celltype(self):
gd_orig = nest.GetDefaults
nest.GetDefaults = Mock(return_value={"FOO": 1.2, "HOO": 3.4, "WOO": 5.6})
p = Population(10, MockNativeCellType)
p.all_cells = numpy.array([MockID()]*10, dtype=object) #numpy.arange(10)
p._mask_local = numpy.ones((10,), bool)
p.set("FOO", 32)
assert_equal(nest.SetStatus.call_args[0][1], {"FOO": 32.0})
p.set("HOO", 33.0)
assert_equal(nest.SetStatus.call_args[0][1], {"HOO": 33.0})
p.set("WOO", 6.0)
assert_equal(nest.SetStatus.call_args[0][1], {"WOO": 6.0})
nest.GetDefaults = gd_orig
def test_set_with_standard_celltype(self):
p = Population(10, MockStandardCellType)
p.all_cells = numpy.array([MockID()] * 10, dtype=object) # numpy.arange(10)
p._mask_local = numpy.ones((10,), bool)
p.set("foo", 32)
assert_equal(nest.SetStatus.call_args[0][1], {"FOO": 32.0})
p.set("hoo", 33.0)
assert_equal(nest.SetStatus.call_args[0][1], {"HOO": 99.0})
p.set("woo", 6.0)
assert_equal(nest.SetStatus.call_args[0][1], {})
p.all_cells[0].set_parameters.assert_called_with(woo=6.0)
def test_set_with_native_celltype(self):
gd_orig = nest.GetDefaults
nest.GetDefaults = Mock(return_value={"FOO": 1.2, "HOO": 3.4, "WOO": 5.6})
p = Population(10, MockNativeCellType)
p.all_cells = numpy.array([MockID()] * 10, dtype=object) # numpy.arange(10)
p._mask_local = numpy.ones((10,), bool)
p.set("FOO", 32)
assert_equal(nest.SetStatus.call_args[0][1], {"FOO": 32.0})
p.set("HOO", 33.0)
assert_equal(nest.SetStatus.call_args[0][1], {"HOO": 33.0})
p.set("WOO", 6.0)
assert_equal(nest.SetStatus.call_args[0][1], {"WOO": 6.0})
nest.GetDefaults = gd_orig
def __init__(self, topic='/laneletInformation'):
super(VehicleLaneAlignmentNetworkIn, self).__init__()
self._vehicle_alignment_pop = Population(1, IF_curr_alpha, {'i_offset': 0, 'v_thresh': 100,
'tau_refrac': 0.1})
detector = nest.nest.Create('spike_detector', params={'withgid': True, 'withtime': True})[0]
self.detectors[self._vehicle_alignment_pop[0]] = detector
nest.nest.Connect(self._vehicle_alignment_pop[0], detector)
self._state = None
self._sub_lanelet = rospy.Subscriber(topic, Float64MultiArray, self._update_state, queue_size=1)
self.output_pop = self._vehicle_alignment_pop
# make sure we get at least one message before starting the training
rospy.wait_for_message(topic, Float64MultiArray)
def _build_input_layer(self):
num_neurons = self._width * self._height
# Layer 1 (Input Image)
self.pop_input_image_r = Population(num_neurons // 2, IF_curr_alpha, {'i_offset': np.zeros(num_neurons // 2)})
self.pop_input_image_l = Population(num_neurons // 2, IF_curr_alpha, {'i_offset': np.zeros(num_neurons // 2)})
# Two 4 x 4 grids, for left and right respectively
# Layer 2: 4 x 4 Grid left
self.pop_in_l2 = Population(4, IF_curr_alpha, {'i_offset': np.zeros(4), 'v_thresh': 100})
conn_list_l = []
for neuron in self.pop_input_image_l:
neuron = neuron - self.pop_input_image_l.first_id
if (neuron % 50) <= 25 and neuron <= 1250: # oben links
conn_list_l.append((neuron, 0, 1.0, 0.1))
if (neuron % 50) > 25 and neuron <= 1250: # oben rechts
conn_list_l.append((neuron, 1, 1.0, 0.1))
if (neuron % 50) <= 25 and neuron > 1250: # unten links
conn_list_l.append((neuron, 2, 1.0, 0.1))
if (neuron % 50) > 25 and neuron > 1250: # unten rechts DOPPELT gewichtet
conn_list_l.append((neuron, 3, 1.0, 0.1))
# Layer 2: 4 x 4 Grid right
self.pop_in_r2 = Population(4, IF_curr_alpha, {'i_offset': np.zeros(4), 'v_thresh': 100})
conn_list_r = []
for neuron in self.pop_input_image_r:
neuron = neuron - self.pop_input_image_r.first_id
if (neuron % 50) <= 25 and neuron <= 1250: # oben links
conn_list_r.append((neuron, 0, 1.0, 0.1))
if (neuron % 50) > 25 and neuron <= 1250: # oben recht
conn_list_r.append((neuron, 1, 1.0, 0.1))
if (neuron % 50) <= 25 and neuron > 1250: # unten links DOPPELT gewichtet
conn_list_r.append((neuron, 2, 1.0, 0.1))
if (neuron % 50) > 25 and neuron > 1250: # unten rechts
conn_list_r.append((neuron, 3, 1.0, 0.1))
# Layer 3 encoded ( The two 4x4 Grids are connected with one neuron each)
self.pop_encoded_image_l = Population(1, IF_curr_alpha, {'tau_refrac': 0.1, 'v_thresh': -50.})
self.pop_encoded_image_r = Population(1, IF_curr_alpha, {'tau_refrac': 0.1, 'v_thresh': -50.})
# Connections: Layer 1 (Input Image) --> Layer 2 (4 x 4 Grids)
self.projection_layer2_l = Projection(self.pop_input_image_l, self.pop_in_l2, FromListConnector(conn_list_l))
self.projection_layer2_r = Projection(self.pop_input_image_r, self.pop_in_r2, FromListConnector(conn_list_r))
self.projection_layer2_l.setWeights(1.0)
self.projection_layer2_r.setWeights(1.0)
# Connections: Layer 2 (4 x 4 Grids) --> Layer 3 (Encoded Image)
# Excitatory connections to "this" side ('left' -> 'left', 'right' -> 'right')
self.projection_out_l = Projection(self.pop_in_l2, self.pop_encoded_image_l, AllToAllConnector())
self.projection_out_r = Projection(self.pop_in_r2, self.pop_encoded_image_r, AllToAllConnector())
self.projection_out_l.setWeights(1.0)
self.projection_out_r.setWeights(1.0)
# bild seiten staerker gewichtet
self.projection_out_l[0]._set_weight(2.0)
self.projection_out_r[1]._set_weight(2.0)
self.projection_out_l[2]._set_weight(4.0)
self.projection_out_r[3]._set_weight(4.0)
# Connections: Layer 2 (4 x 4 Grids) --> Layer 3 (Encoded Image)
# Inhibitory connections to "other" side ('left' -> 'right', 'right' -> 'left')
self.projection_out_l_i = Projection(self.pop_in_l2, self.pop_encoded_image_r, AllToAllConnector())
self.projection_out_r_i = Projection(self.pop_in_r2, self.pop_encoded_image_l, AllToAllConnector())
self.projection_out_l_i.setWeights(-1)
self.projection_out_r_i.setWeights(-1)
self.output_pop = [self.pop_encoded_image_l, self.pop_encoded_image_r]
class BaseNetworkIn(BaseNetwork):
""" Handles the frame injection and provides two input neurons (left, right)"""
@property
def last_frame(self):
return self._last_frame
@property
def encoded_image_pops(self):
return {'left': self.pop_encoded_image_l, 'right': self.pop_encoded_image_r}
def __init__(self, image_topic='/spiky/retina_image'):
super(BaseNetworkIn, self).__init__()
self._postsynaptic_learning_neurons = []
# Populations
self.pop_input_image_r = None
self.pop_input_image_l = None
self.pop_encoded_image_l = None
self.pop_encoded_image_r = None
self.spikedetector_enc_image_l = None
self.spikedetector_enc_image_r = None
# Preparing sensory information subscriber
self._bridge = CvBridge()
self._last_frame = None
self._retina_image_subscriber = rospy.Subscriber(image_topic, Image, self._handle_frame, queue_size=1)
rospy.wait_for_message(image_topic, Image)
if self._last_frame is not None:
shape = np.shape(self._last_frame)
self._width = shape[0]
self._height = shape[1]
self._build_input_layer()
self._create_spike_detectors()
def _build_input_layer(self):
num_neurons = self._width * self._height
# Layer 1 (Input Image)
self.pop_input_image_r = Population(num_neurons // 2, IF_curr_alpha, {'i_offset': np.zeros(num_neurons // 2)})
self.pop_input_image_l = Population(num_neurons // 2, IF_curr_alpha, {'i_offset': np.zeros(num_neurons // 2)})
# Two 4 x 4 grids, for left and right respectively
# Layer 2: 4 x 4 Grid left
self.pop_in_l2 = Population(4, IF_curr_alpha, {'i_offset': np.zeros(4), 'v_thresh': 100})
conn_list_l = []
for neuron in self.pop_input_image_l:
neuron = neuron - self.pop_input_image_l.first_id
if (neuron % 50) <= 25 and neuron <= 1250: # oben links
conn_list_l.append((neuron, 0, 1.0, 0.1))
if (neuron % 50) > 25 and neuron <= 1250: # oben rechts
conn_list_l.append((neuron, 1, 1.0, 0.1))
if (neuron % 50) <= 25 and neuron > 1250: # unten links
conn_list_l.append((neuron, 2, 1.0, 0.1))
if (neuron % 50) > 25 and neuron > 1250: # unten rechts DOPPELT gewichtet
conn_list_l.append((neuron, 3, 1.0, 0.1))
# Layer 2: 4 x 4 Grid right
self.pop_in_r2 = Population(4, IF_curr_alpha, {'i_offset': np.zeros(4), 'v_thresh': 100})
conn_list_r = []
for neuron in self.pop_input_image_r:
neuron = neuron - self.pop_input_image_r.first_id
if (neuron % 50) <= 25 and neuron <= 1250: # oben links
conn_list_r.append((neuron, 0, 1.0, 0.1))
if (neuron % 50) > 25 and neuron <= 1250: # oben recht
conn_list_r.append((neuron, 1, 1.0, 0.1))
if (neuron % 50) <= 25 and neuron > 1250: # unten links DOPPELT gewichtet
conn_list_r.append((neuron, 2, 1.0, 0.1))
if (neuron % 50) > 25 and neuron > 1250: # unten rechts
conn_list_r.append((neuron, 3, 1.0, 0.1))
# Layer 3 encoded ( The two 4x4 Grids are connected with one neuron each)
self.pop_encoded_image_l = Population(1, IF_curr_alpha, {'tau_refrac': 0.1, 'v_thresh': -50.})
self.pop_encoded_image_r = Population(1, IF_curr_alpha, {'tau_refrac': 0.1, 'v_thresh': -50.})
# Connections: Layer 1 (Input Image) --> Layer 2 (4 x 4 Grids)
self.projection_layer2_l = Projection(self.pop_input_image_l, self.pop_in_l2, FromListConnector(conn_list_l))
self.projection_layer2_r = Projection(self.pop_input_image_r, self.pop_in_r2, FromListConnector(conn_list_r))
self.projection_layer2_l.setWeights(1.0)
self.projection_layer2_r.setWeights(1.0)
# Connections: Layer 2 (4 x 4 Grids) --> Layer 3 (Encoded Image)
# Excitatory connections to "this" side ('left' -> 'left', 'right' -> 'right')
self.projection_out_l = Projection(self.pop_in_l2, self.pop_encoded_image_l, AllToAllConnector())
self.projection_out_r = Projection(self.pop_in_r2, self.pop_encoded_image_r, AllToAllConnector())
self.projection_out_l.setWeights(1.0)
self.projection_out_r.setWeights(1.0)
# bild seiten staerker gewichtet
self.projection_out_l[0]._set_weight(2.0)
self.projection_out_r[1]._set_weight(2.0)
self.projection_out_l[2]._set_weight(4.0)
self.projection_out_r[3]._set_weight(4.0)
# Connections: Layer 2 (4 x 4 Grids) --> Layer 3 (Encoded Image)
# Inhibitory connections to "other" side ('left' -> 'right', 'right' -> 'left')
self.projection_out_l_i = Projection(self.pop_in_l2, self.pop_encoded_image_r, AllToAllConnector())
self.projection_out_r_i = Projection(self.pop_in_r2, self.pop_encoded_image_l, AllToAllConnector())
self.projection_out_l_i.setWeights(-1)
self.projection_out_r_i.setWeights(-1)
self.output_pop = [self.pop_encoded_image_l, self.pop_encoded_image_r]
def _handle_frame(self, frame):
self._last_frame = self._bridge.imgmsg_to_cv2(frame)
def _create_spike_detectors(self):
self.spikedetector_enc_image_l = nest.nest.Create('spike_detector', params={'withgid': True, 'withtime': True})[
0]
self.spikedetector_enc_image_r = nest.nest.Create('spike_detector', params={'withgid': True, 'withtime': True})[
0]
nest.nest.Connect(self.pop_encoded_image_l[0], self.spikedetector_enc_image_l)
nest.nest.Connect(self.pop_encoded_image_r[0], self.spikedetector_enc_image_r)
self.detectors[self.pop_encoded_image_l[0]] = self.spikedetector_enc_image_l
self.detectors[self.pop_encoded_image_r[0]] = self.spikedetector_enc_image_r
def before_simulation(self):
self.inject_frame(self.last_frame)
def inject_frame(self, frame):
frame_l = frame[0:50, 0:50]
frame_r = frame[0:50, 50:100]
self.pop_input_image_l.set(i_offset=frame_l.astype(float).flatten())
self.pop_input_image_r.set(i_offset=frame_r.astype(float).flatten())
def populate_plotter(self, plotter):
plotter.add_spike_train_plot(self.pop_encoded_image_l, label='Input Left')
plotter.add_spike_train_plot(self.pop_encoded_image_r, label='Input Right')
import pyNN
from pyNN.nest import Population, SpikeSourcePoisson, SpikeSourceArray, AllToAllConnector, run, setup, IF_curr_alpha
from pyNN.nest.projections import Projection
import matplotlib.pyplot as plt
setup(timestep=0.1)
image = cv2.imread('/fzi/ids/mlprak2/Bilder/impuls.tif')
image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
frame = np.zeros(image.size) + 255
rates_init = np.zeros(frame.size)
pop_in = Population(frame.shape, SpikeSourcePoisson, {'rate': rates_init})
pop_out = Population(1, IF_curr_alpha, {'tau_refrac': 5 })
projection = Projection(pop_in, pop_out, AllToAllConnector())
projection.setWeights(1.0)
pop_in.set(rate=image.astype(float).flatten())
pop_in.record('spikes')
pop_out.record('spikes')
tstop = 100.0
run(tstop)
spikes_in = pop_in.get_data()