nengo.Connection(targetVelY,currPosY,transform=tau,synapse=tau/10)
nengo.Connection(currPosY,currPosY,synapse=tau)
synP = synC #0.01
prb_ip1 = nengo.Probe(ip1,synapse=0.01)
prb_ip2 = nengo.Probe(ip2,synapse=0.01)
prb_posX = nengo.Probe(currPosX,synapse=synP)
prb_posY = nengo.Probe(currPosY,synapse=synP)
prb_velX = nengo.Probe(targetVelX,synapse=synP)
prb_velY = nengo.Probe(targetVelY,synapse=synP)
prb_posXn = nengo.Probe(currPosX.neurons)
prb_posYn = nengo.Probe(currPosY.neurons)
prb_velXn = nengo.Probe(targetVelX.neurons)
prb_velYn = nengo.Probe(targetVelY.neurons)
with nengo.Simulator(eye_ctrl) as sim:
sim.run(5)
t = sim.trange()
plt.figure()
plt.plot(t,sim.data[prb_posX],label='posX')
plt.plot(t,sim.data[prb_posY],label='posX')
plt.plot(t,sim.data[prb_ip1],label='inputX')
plt.plot(t,sim.data[prb_ip2],label='inputY')
plt.xlabel("time(s)")
plt.ylabel("rad")
plt.title("Eye Position")
plt.figure()
plt.plot(t,sim.data[prb_posX],label='posX')
def test_conv_connection(channels, Simulator, seed, rng, plt, allclose):
# channels_last = True
channels_last = False
if channels > 1:
pytest.xfail("Cannot send population spikes to chip")
# load data
with open(os.path.join(test_dir, 'mnist10.pkl'), 'rb') as f:
test10 = pickle.load(f)
test_x = test10[0][0].reshape(28, 28)
test_x = 1.999 * test_x - 0.999 # range (-1, 1)
test_x = test_x[:, :, None] # single channel
input_shape = ImageShape(test_x.shape[0],
test_x.shape[1],
channels,
channels_last=channels_last)
filters = Gabor(freq=Uniform(0.5, 1)).generate(8, (7, 7), rng=rng)
filters = filters[None, :, :, :] # single channel
filters = np.transpose(filters, (0, 2, 3, 1)) # filters last
strides = (2, 2)
tau_rc = 0.02
tau_ref = 0.002
tau_s = 0.005
dt = 0.001
neuron_type = LoihiLIF(tau_rc=tau_rc, tau_ref=tau_ref)
pres_time = 1.0
with nengo.Network(seed=seed) as model:
nengo_loihi.add_params(model)
u = nengo.Node(nengo.processes.PresentInput([test_x.ravel()],
pres_time),
label='u')
a = nengo.Ensemble(input_shape.size,
1,
neuron_type=LoihiSpikingRectifiedLinear(),
max_rates=nengo.dists.Choice([40 / channels]),
intercepts=nengo.dists.Choice([0]),
label='a')
model.config[a].on_chip = False
if channels == 1:
nengo.Connection(u, a.neurons, transform=1, synapse=None)
elif channels == 2:
# encode image into spikes using two channels (on/off)
if input_shape.channels_last:
nengo.Connection(u, a.neurons[0::2], transform=1, synapse=None)
nengo.Connection(u,
a.neurons[1::2],
transform=-1,
synapse=None)
else:
k = input_shape.rows * input_shape.cols
nengo.Connection(u, a.neurons[:k], transform=1, synapse=None)
nengo.Connection(u, a.neurons[k:], transform=-1, synapse=None)
filters = np.vstack([filters, -filters])
else:
raise ValueError("Test not configured for more than two channels")
conv2d_transform = Conv2D.from_kernel(filters,
input_shape,
strides=strides)
output_shape = conv2d_transform.output_shape
gain, bias = neuron_type.gain_bias(max_rates=100, intercepts=0)
gain = gain * 0.01 # account for `a` max_rates
b = nengo.Ensemble(output_shape.size,
1,
neuron_type=neuron_type,
gain=nengo.dists.Choice([gain[0]]),
bias=nengo.dists.Choice([bias[0]]),
label='b')
nengo.Connection(a.neurons,
b.neurons,
synapse=tau_s,
transform=conv2d_transform)
bp = nengo.Probe(b.neurons)
with nengo.Simulator(model, dt=dt, optimize=False) as sim:
sim.run(pres_time)
ref_out = sim.data[bp].mean(axis=0).reshape(output_shape.shape())
# Currently, default TensorFlow does not support channels first in conv
use_nengo_dl = nengo_dl is not None and channels_last
ndl_out = np.zeros_like(ref_out)
if use_nengo_dl:
with nengo_dl.Simulator(model, dt=dt) as sim:
sim.run(pres_time)
ndl_out = sim.data[bp].mean(axis=0).reshape(output_shape.shape())
with nengo_loihi.Simulator(model, dt=dt, target='simreal') as sim:
sim.run(pres_time)
real_out = sim.data[bp].mean(axis=0).reshape(output_shape.shape())
with Simulator(model, dt=dt) as sim:
sim.run(pres_time)
sim_out = sim.data[bp].mean(axis=0).reshape(output_shape.shape())
if not output_shape.channels_last:
ref_out = np.transpose(ref_out, (1, 2, 0))
ndl_out = np.transpose(ndl_out, (1, 2, 0))
real_out = np.transpose(real_out, (1, 2, 0))
sim_out = np.transpose(sim_out, (1, 2, 0))
out_max = max(ref_out.max(), sim_out.max())
# --- plot results
rows = 2
cols = 3
ax = plt.subplot(rows, cols, 1)
imshow(test_x, vmin=0, vmax=1, ax=ax)
ax = plt.subplot(rows, cols, 2)
tile(np.transpose(filters[0], (2, 0, 1)), cols=8, ax=ax)
ax = plt.subplot(rows, cols, 3)
plt.hist(ref_out.ravel(), bins=31)
plt.hist(sim_out.ravel(), bins=31)
ax = plt.subplot(rows, cols, 4)
tile(np.transpose(ref_out, (2, 0, 1)), vmin=0, vmax=out_max, cols=8, ax=ax)
ax = plt.subplot(rows, cols, 5)
tile(np.transpose(ndl_out, (2, 0, 1)), vmin=0, vmax=out_max, cols=8, ax=ax)
ax = plt.subplot(rows, cols, 6)
tile(np.transpose(sim_out, (2, 0, 1)), vmin=0, vmax=out_max, cols=8, ax=ax)
if use_nengo_dl:
assert allclose(ndl_out, ref_out, atol=1e-5, rtol=1e-5)
assert allclose(real_out, ref_out, atol=1, rtol=1e-3)
assert allclose(sim_out, ref_out, atol=10, rtol=1e-3)
def evaluate_mnist_multiple_baseline_noise(args):
#############################
# load the data
#############################
input_nbr = args.input_nbr
input_nbr = args.input_nbr
probe_sample_rate = (input_nbr/10)/1000 #Probe sample rate. Proportional to input_nbr to scale down sampling rate of simulations
x = args.digit
np.random.seed(args.seed)
random.seed(args.seed)
data = np.load('mnist_norm.npz', allow_pickle=True)
image_train_filtered = data['image_train_filtered']/255
label_train_filtered = data['label_train_filtered']
image_test_filtered = data['image_test_filtered']/255
label_test_filtered = data['label_test_filtered']
image_assign_filtered = image_train_filtered
label_assign_filtered = label_train_filtered
image_train_filtered = np.tile(image_train_filtered,(args.iterations,1,1))
label_train_filtered = np.tile(label_train_filtered,(args.iterations))
#Simulation Parameters
#Presentation time
presentation_time = args.presentation_time #0.20
#Pause time
# pause_time = args.pause_time + 0.0001
pause_time = args.pause_time
#Iterations
iterations=args.iterations
#Input layer parameters
n_in = args.n_in
# g_max = 1/784 #Maximum output contribution
amp_neuron = args.amp_neuron
n_neurons = args.n_neurons # Layer 1 neurons
# inhib_factor = args.inhib_factor #Multiplication factor for lateral inhibition
input_neurons_args = {
"n_neurons":n_in,
"dimensions":1,
"label":"Input layer",
"encoders":nengo.dists.Choice([[1]]),
# "max_rates":nengo.dists.Uniform(22,22),
# "intercepts":nengo.dists.Uniform(0,0),
"gain":nengo.dists.Choice([args.gain_in]),
"bias":nengo.dists.Choice([args.bias_in]),
"noise":nengo.processes.WhiteNoise(dist=nengo.dists.Gaussian(0, args.noise_input), seed=args.seed),
"neuron_type":MyLIF_in_v2(tau_rc=args.tau_in,min_voltage=-1, amplitude=args.amp_neuron, tau_ref=args.tau_ref_in)
# "neuron_type":nengo.neurons.SpikingRectifiedLinear()#SpikingRelu neuron.
}
#Layer 1 parameters
layer_1_neurons_args = {
"n_neurons":n_neurons,
"dimensions":1,
"label":"Layer 1",
"encoders":nengo.dists.Choice([[1]]),
"gain":nengo.dists.Choice([args.gain_out]),
"bias":nengo.dists.Choice([args.bias_out]),
# "intercepts":nengo.dists.Choice([0]),
# "max_rates":nengo.dists.Choice([args.rate_out,args.rate_out]),
# "noise":nengo.processes.WhiteNoise(dist=nengo.dists.Gaussian(0, 0.5), seed=1),
# "neuron_type":nengo.neurons.LIF(tau_rc=args.tau_out, min_voltage=0)
# "neuron_type":MyLIF_out(tau_rc=args.tau_out, min_voltage=-1)
"neuron_type":STDPLIF(tau_rc=args.tau_out, min_voltage=-1, spiking_threshold=args.thr_out, inhibition_time=args.inhibition_time,tau_ref=args.tau_ref_out,inc_n=args.inc_n,tau_n=args.tau_n)
}
#Learning rule parameters
learning_args = {
"lr": args.lr,
"alpha": args.alpha,
"winit_min":0,
"winit_max":args.winit_max,
"sample_distance": int((presentation_time+pause_time)*200*10), #Store weight after 10 images
}
# argument_string = "presentation_time: "+ str(presentation_time)+ "\n pause_time: "+ str(pause_time)+ "\n input_neurons_args: " + str(input_neurons_args)+ " \n layer_1_neuron_args: " + str(layer_1_neurons_args)+"\n Lateral Inhibition parameters: " + str(lateral_inhib_args) + "\n learning parameters: " + str(learning_args)+ "\n g_max: "+ str(g_max)
images = image_train_filtered
labels = label_train_filtered
np.random.seed(args.seed)
random.seed(args.seed)
model = nengo.Network("My network", seed = args.seed)
#############################
# Model construction
#############################
with model:
# picture = nengo.Node(PresentInputWithPause(images, presentation_time, pause_time,0))
picture = nengo.Node(nengo.processes.PresentInput(images, presentation_time=presentation_time))
true_label = nengo.Node(nengo.processes.PresentInput(labels, presentation_time=presentation_time))
# true_label = nengo.Node(PresentInputWithPause(labels, presentation_time, pause_time,-1))
# input layer
input_layer = nengo.Ensemble(**input_neurons_args)
input_conn = nengo.Connection(picture,input_layer.neurons,synapse=None)
#first layer
layer1 = nengo.Ensemble(**layer_1_neurons_args)
#Weights between input layer and layer 1
w = nengo.Node(CustomRule_post_baseline(**learning_args), size_in=n_in, size_out=n_neurons)
nengo.Connection(input_layer.neurons, w, synapse=None)
nengo.Connection(w, layer1.neurons, synapse=args.synapse_layer_1)
weights = w.output.history
# with nengo_ocl.Simulator(model) as sim :
with nengo.Simulator(model, dt=args.dt, optimize=True) as sim:
w.output.set_signal_vmem(sim.signals[sim.model.sig[input_layer.neurons]["voltage"]])
w.output.set_signal_out(sim.signals[sim.model.sig[layer1.neurons]["out"]])
sim.run((presentation_time+pause_time) * labels.shape[0])
last_weight = weights[-1]
sim.close()
pause_time = 0
#Neuron class assingment
images = image_assign_filtered
labels = label_assign_filtered
model = nengo.Network("My network", seed = args.seed)
with model:
# picture = nengo.Node(PresentInputWithPause(images, presentation_time, pause_time,0))
picture = nengo.Node(nengo.processes.PresentInput(images, presentation_time=presentation_time))
true_label = nengo.Node(nengo.processes.PresentInput(labels, presentation_time=presentation_time))
# true_label = nengo.Node(PresentInputWithPause(labels, presentation_time, pause_time,-1))
# input layer
input_layer = nengo.Ensemble(**input_neurons_args)
input_conn = nengo.Connection(picture,input_layer.neurons,synapse=None)
#first layer
layer1 = nengo.Ensemble(**layer_1_neurons_args)
nengo.Connection(input_layer.neurons, layer1.neurons,transform=last_weight,synapse=args.synapse_layer_1)
#Probes
p_true_label = nengo.Probe(true_label)
p_layer_1 = nengo.Probe(layer1.neurons)
# with nengo_ocl.Simulator(model) as sim :
with nengo.Simulator(model, dt=args.dt, optimize=True) as sim:
sim.run((presentation_time+pause_time) * labels.shape[0])
t_data = sim.trange()
labels = sim.data[p_true_label][:,0]
output_spikes = sim.data[p_layer_1]
neuron_class = np.zeros((n_neurons, 1))
n_classes = 10
for j in range(n_neurons):
spike_times_neuron_j = t_data[np.where(output_spikes[:,j] > 0)]
max_spike_times = 0
for i in range(n_classes):
class_presentation_times_i = t_data[np.where(labels == i)]
#Normalized number of spikes wrt class presentation time
num_spikes = len(np.intersect1d(spike_times_neuron_j,class_presentation_times_i))/(len(class_presentation_times_i)+1)
if(num_spikes>max_spike_times):
neuron_class[j] = i
max_spike_times = num_spikes
spikes_layer1_probe_train = sim.data[p_layer_1]
#Testing
images = image_test_filtered
labels = label_test_filtered
input_nbr = 10000
model = nengo.Network(label="My network",)
with model:
# picture = nengo.Node(PresentInputWithPause(images, presentation_time, pause_time,0))
picture = nengo.Node(nengo.processes.PresentInput(images, presentation_time=presentation_time))
true_label = nengo.Node(nengo.processes.PresentInput(labels, presentation_time=presentation_time))
# true_label = nengo.Node(PresentInputWithPause(labels, presentation_time, pause_time,-1))
input_layer = nengo.Ensemble(**input_neurons_args)
input_conn = nengo.Connection(picture,input_layer.neurons,synapse=None)
#first layer
layer1 = nengo.Ensemble(**layer_1_neurons_args)
nengo.Connection(input_layer.neurons, layer1.neurons,transform=last_weight,synapse=args.synapse_layer_1)
p_true_label = nengo.Probe(true_label)
p_layer_1 = nengo.Probe(layer1.neurons)
step_time = (presentation_time + pause_time)
with nengo.Simulator(model,dt=args.dt) as sim:
sim.run(presentation_time * label_test_filtered.shape[0])
accuracy_2 = evaluation_v2(10,n_neurons,int(((presentation_time * label_test_filtered.shape[0]) / sim.dt) / input_nbr),spikes_layer1_probe_train,label_train_filtered,sim.data[p_layer_1],label_test_filtered,sim.dt)
labels = sim.data[p_true_label][:,0]
t_data = sim.trange()
output_spikes = sim.data[p_layer_1]
n_classes = 10
predicted_labels = []
true_labels = []
correct_classified = 0
wrong_classified = 0
class_spikes = np.ones((10,1))
for num in range(input_nbr):
#np.sum(sim.data[my_spike_probe] > 0, axis=0)
output_spikes_num = output_spikes[num*int((presentation_time + pause_time) /args.dt):(num+1)*int((presentation_time + pause_time) /args.dt),:] # 0.350/0.005
num_spikes = np.sum(output_spikes_num > 0, axis=0)
for i in range(n_classes):
sum_temp = 0
count_temp = 0
for j in range(n_neurons):
if((neuron_class[j]) == i) :
sum_temp += num_spikes[j]
count_temp +=1
if(count_temp==0):
class_spikes[i] = 0
else:
class_spikes[i] = sum_temp
# class_spikes[i] = sum_temp/count_temp
# print(class_spikes)
k = np.argmax(num_spikes)
# predicted_labels.append(neuron_class[k])
class_pred = np.argmax(class_spikes)
predicted_labels.append(class_pred)
true_class = labels[(num*int((presentation_time + pause_time) /args.dt))]
if(class_pred == true_class):
correct_classified+=1
else:
wrong_classified+=1
accuracy = correct_classified/ (correct_classified+wrong_classified)*100
print("Accuracy: ", accuracy)
sim.close()
del sim.data, labels, class_pred, spikes_layer1_probe_train
return accuracy, accuracy_2, weights[-1]
def model_init(self):
model = nengo.Network(label='Neuron Scheduler')
with model:
# running_processes = waiting_processes
print(self.running_process_size())
num_cores_node = nengo.Node(self.num_cores)
running_size = nengo.Node(output=self.running_process_size()
) # output=running_process_size())
running_procs_size_en = nengo.Ensemble(self.calc_n_neurons,
1,
radius=self.num_cores)
num_cores_en = nengo.Ensemble(1024, 1, radius=self.num_cores)
nengo.Connection(num_cores_node, num_cores_en)
avail_procs = nengo.Ensemble(self.calc_n_neurons,
1,
radius=self.num_cores)
nengo.Connection(num_cores_en, avail_procs)
# nengo.Connection(running_procs_size_en, avail_procs.neurons, transform=[[-1]]*calc_n_neurons)
nengo.Connection(running_procs_size_en, avail_procs, transform=-1)
def can_add_thd_logic(x):
x = np.round(x)
return x
## New Process Addition
waiting_proc_q_top_size = nengo.Node(1)
waiting_proc_size = nengo.Ensemble(self.calc_n_neurons,
1,
radius=self.num_cores)
def waiting_q_input_logic(x):
if x <= 0:
return self.num_cores * 2
else:
return x
# nengo.Connection(waiting_proc_q_top_size,waiting_proc_size,function=waiting_q_input_logic)
can_add_next_proc = nengo.Ensemble(self.calc_n_neurons, 1)
next_proc_size_check = nengo.Ensemble(
self.calc_n_neurons,
1,
radius=self.num_cores,
neuron_type=nengo.neurons.SpikingRectifiedLinear())
next_proc_size_check.intercepts = Choice([-.1])
nengo.Connection(avail_procs, next_proc_size_check)
nengo.Connection(waiting_proc_size,
next_proc_size_check,
transform=-1)
def can_add_next_proc_logic(x):
if x > 0:
return 1
else:
return 0
nengo.Connection(next_proc_size_check,
can_add_next_proc,
function=can_add_thd_logic)
### Add New
## Queue Repr:
qsim_waiting = nengo.Node(self.queue_nodes.output_message,
size_out=3)
qsim_add_spike = nengo.Node(self.queue_nodes.input_message,
size_in=1,
size_out=1)
v = nengo.Ensemble(32, 3, radius=128)
nengo.Connection(qsim_waiting, v)
nengo.Connection(qsim_waiting[0],
waiting_proc_size,
function=waiting_q_input_logic)
# nengo.Connection(qsim_waiting[1] , running_procs_size_en)
add_new_proc_system = nengo.Ensemble(
self.calc_n_neurons,
1,
radius=2,
neuron_type=nengo.Izhikevich(reset_recovery=0))
nengo.Connection(can_add_next_proc,
add_new_proc_system,
function=np.ceil)
nengo.Connection(running_size, running_procs_size_en)
## Filter for spikes:
nengo.Connection(add_new_proc_system, qsim_add_spike)
# deepcode ignore MultiplyList: Pointing to the list is okay here since neurons are read-only
nengo.Connection(qsim_add_spike,
add_new_proc_system.neurons,
transform=[[-1]] * self.calc_n_neurons)
## RR Clock
clock_input = nengo.Node(size_out=1)
check_proc_ens = nengo.Ensemble(self.calc_n_neurons, dimensions=2)
rr_clock = nengo.Network()
with rr_clock:
def node_tick_clock(t):
t = np.round(t) % 10
return t
def neuron_timer_check(x):
nv = x[0]
if x[0] > 10:
nv = 0
else:
nv = x[0]
return nv
stim = nengo.Node(1)
inhibition = nengo.Node(0)
interupt_proc = nengo.Ensemble(n_neurons=1024, dimensions=1)
c = nengo.Ensemble(n_neurons=1024, dimensions=1)
nengo.Connection(stim, interupt_proc)
nengo.Connection(c,
interupt_proc.neurons,
transform=[[-2.5]] * 1024)
nengo.Connection(inhibition, c)
clock = nengo.Node(node_tick_clock)
time_slice = nengo.Node(10)
cl_enc = nengo.Ensemble(1024, dimensions=1, radius=25)
ts_enc = nengo.Ensemble(1024, dimensions=1, radius=25)
nengo.Connection(time_slice,
ts_enc,
function=lambda x: x % self.time_slice_ticks)
nengo.Connection(clock, cl_enc)
time_check = nengo.Ensemble(1024, dimensions=1, radius=20)
# summation = nengo.networks.
tau = 0.1
nengo.Connection(cl_enc,
time_check,
synapse=tau,
function=neuron_timer_check)
nengo.Connection(time_check, c)
qsim_running_interrupt = nengo.Node(size_in=1, size_out=0)
nengo.Connection(interupt_proc, qsim_running_interrupt)
self.model = model
self.sim = nengo.Simulator(model)
stim_blue = nengo.Node(Piecewise(dict(zip(time, input))))
red = gbellmf([0.25, 8, -0.6])
nengo.Connection(stim_red, red.I)
green = gbellmf([0.35, 2, -0.1])
nengo.Connection(stim_green, green.I)
blue = gbellmf([0.25, 3, 0.3])
nengo.Connection(stim_blue, blue.I)
with model:
pr_red = nengo.Probe(red.O, synapse=0.01)
pr_green = nengo.Probe(green.O, synapse=0.01)
pr_blue = nengo.Probe(blue.O, synapse=0.01)
with nengo.Simulator(model) as sim:
sim.run(2.5)
t = sim.trange()
print(len(sim.data[pr_red]))
plt.figure()
plt.plot(t[500:] - 1.5, sim.data[pr_red][500:], c='r', label="Red")
plt.plot(t[500:] - 1.5, sim.data[pr_green][500:], c='g', label="Green")
plt.plot(t[500:] - 1.5, sim.data[pr_blue][500:], c='b', label="Blue")
plt.xlim(right=1)
plt.xlabel("Input")
plt.ylabel("Degree of membership")
plt.legend(loc="best")
plt.show()
def run_nn(simulator=None, model=None, time=10):
sim = simulator
if sim == None:
sim = nengo.Simulator(model)
sim.run(time)
return [model, sim]
with srf_network:
p_output_spikes = nengo.Probe(srf_network.output.neurons)
p_inh_weights = nengo.Probe(srf_network.conn_I,
'weights',
sample_every=1.0)
p_exc_weights = nengo.Probe(srf_network.conn_E,
'weights',
sample_every=1.0)
p_rec_weights = nengo.Probe(srf_network.conn_EE,
'weights',
sample_every=1.0)
p_exc_rates = nengo.Probe(srf_network.exc.neurons)
p_inh_rates = nengo.Probe(srf_network.inh.neurons)
with nengo.Simulator(srf_network, optimize=True) as sim:
sim.run(0.01)
exc_weights = srf_network.conn_E.transform
inh_weights = srf_network.conn_I.transform
rec_weights = srf_network.conn_EE.transform
plt.imshow(np.asarray(exc_weights.sample()).T,
aspect="auto",
interpolation="nearest")
plt.colorbar()
plt.show()
plt.imshow(np.asarray(rec_weights.sample()).T,
aspect="auto",
interpolation="nearest")
plt.colorbar()
def evaluate_mnist_single(args):
#############################
# load the data
#############################
input_nbr = args.input_nbr
(image_train, label_train), (image_test, label_test) = (tf.keras.datasets.fashion_mnist.load_data())
probe_sample_rate = (input_nbr/10)/1000 #Probe sample rate. Proportional to input_nbr to scale down sampling rate of simulations
# probe_sample_rate = 1000
image_train_filtered = []
label_train_filtered = []
x = args.digit
for i in range(0,input_nbr):
if label_train[i] == x:
image_train_filtered.append(image_train[i])
label_train_filtered.append(label_train[i])
image_train_filtered = np.array(image_train_filtered)
label_train_filtered = np.array(label_train_filtered)
#Simulation Parameters
#Presentation time
presentation_time = args.presentation_time #0.20
#Pause time
pause_time = args.pause_time
#Iterations
iterations=args.iterations
#Input layer parameters
n_in = args.n_in
# g_max = 1/784 #Maximum output contribution
g_max = args.g_max
n_neurons = args.n_neurons # Layer 1 neurons
inhib_factor = -0*100 #Multiplication factor for lateral inhibition
n_neurons = 1
input_neurons_args = {
"n_neurons":n_in,
"dimensions":1,
"label":"Input layer",
"encoders":nengo.dists.Uniform(1,1),
"gain":nengo.dists.Uniform(2,2),
"bias":nengo.dists.Uniform(0,0),
"neuron_type":MyLIF_in(tau_rc=args.tau_in,min_voltage=-1, amplitude=args.g_max)
# "neuron_type":nengo.neurons.LIF(tau_rc=args.tau_in,min_voltage=0)#SpikingRelu neuron.
}
#Layer 1 parameters
layer_1_neurons_args = {
"n_neurons":n_neurons,
"dimensions":1,
"label":"Layer 1",
"encoders":nengo.dists.Uniform(1,1),
# "gain":nengo.dists.Uniform(2,2),
# "bias":nengo.dists.Uniform(0,0),
"intercepts":nengo.dists.Choice([0]),
"max_rates":nengo.dists.Choice([20,20]),
"noise":nengo.processes.WhiteNoise(dist=nengo.dists.Gaussian(0, 1), seed=1),
# "neuron_type":nengo.neurons.LIF(tau_rc=args.tau_out, min_voltage=0)
# "neuron_type":MyLIF_out(tau_rc=args.tau_out, min_voltage=-1)
"neuron_type":STDPLIF(tau_rc=args.tau_out, min_voltage=-1),
}
# "noise":nengo.processes.WhiteNoise(dist=nengo.dists.Gaussian(0, 20), seed=1),
#Lateral Inhibition parameters
lateral_inhib_args = {
"transform": inhib_factor* (np.full((n_neurons, n_neurons), 1) - np.eye(n_neurons)),
"synapse":0.01,
"label":"Lateral Inhibition"
}
#Learning rule parameters
learning_args = {
"lr": args.lr,
"winit_min":0,
"winit_max":0.1,
# "tpw":50,
# "prev_flag":True,
"sample_distance": int((presentation_time+pause_time)*200),
}
argument_string = "presentation_time: "+ str(presentation_time)+ "\n pause_time: "+ str(pause_time)+ "\n input_neurons_args: " + str(input_neurons_args)+ " \n layer_1_neuron_args: " + str(layer_1_neurons_args)+"\n Lateral Inhibition parameters: " + str(lateral_inhib_args) + "\n learning parameters: " + str(learning_args)+ "\n g_max: "+ str(g_max)
images = image_train_filtered
labels = label_train_filtered
model = nengo.Network("My network")
#############################
# Model construction
#############################
with model:
picture = nengo.Node(nengo.processes.PresentInput(images, presentation_time))
true_label = nengo.Node(nengo.processes.PresentInput(labels, presentation_time))
# picture = nengo.Node(PresentInputWithPause(images, presentation_time, pause_time,0))
# true_label = nengo.Node(PresentInputWithPause(labels, presentation_time, pause_time,-1))
# input layer
input_layer = nengo.Ensemble(**input_neurons_args)
input_conn = nengo.Connection(picture,input_layer.neurons,synapse=None)
#first layer
layer1 = nengo.Ensemble(**layer_1_neurons_args)
#Weights between input layer and layer 1
w = nengo.Node(CustomRule_post_v2(**learning_args), size_in=n_in, size_out=n_neurons)
nengo.Connection(input_layer.neurons, w, synapse=None)
nengo.Connection(w, layer1.neurons, synapse=None)
#Lateral inhibition
# inhib = nengo.Connection(layer1.neurons,layer1.neurons,**lateral_inhib_args)
#Probes
# p_true_label = nengo.Probe(true_label, sample_every=probe_sample_rate)
p_input_layer = nengo.Probe(input_layer.neurons, sample_every=probe_sample_rate)
p_layer_1 = nengo.Probe(layer1.neurons, sample_every=probe_sample_rate)
weights = w.output.history
with nengo.Simulator(model, dt=0.005) as sim :
# with nengo_spinnaker.Simulator(model) as sim:
w.output.set_signal_vmem(sim.signals[sim.model.sig[input_layer.neurons]["voltage"]])
w.output.set_signal_out(sim.signals[sim.model.sig[layer1.neurons]["out"]])
sim.run((presentation_time+pause_time) * labels.shape[0]*iterations)
#save the model
# now = time.strftime("%Y%m%d-%H%M%S")
# folder = os.getcwd()+"/MNIST_VDSP"+now
# os.mkdir(folder)
last_weight = weights[-1]
# pickle.dump(weights, open( folder+"/trained_weights", "wb" ))
# pickle.dump(argument_string, open( folder+"/arguments", "wb" ))
sim.close()
return weights
def go(NPre=100, N=100, t=10, c=None, seed=0, dt=0.001, Tff=0.3, tTrans=0.01,
stage=None, alpha=3e-7, eMax=1e-1,
fPre=DoubleExp(1e-3, 1e-1), fNMDA=DoubleExp(10.6e-3, 285e-3), fS=DoubleExp(1e-3, 1e-1),
dPreA=None, dPreB=None, dPreC=None, dFdfw=None, dBio=None, dNeg=None, dInh=None,
ePreA=None, ePreB=None, ePreC=None, eFdfw=None, eBio=None, eNeg=None, eInh=None,
stimA=lambda t: 0, stimB=lambda t: 0, stimC=lambda t: 0, DA=lambda t: 0):
if not c: c = t
with nengo.Network(seed=seed) as model:
inptA = nengo.Node(stimA)
inptB = nengo.Node(stimB)
inptC = nengo.Node(stimC)
preA = nengo.Ensemble(NPre, 1, max_rates=Uniform(30, 30), seed=seed)
preB = nengo.Ensemble(NPre, 1, max_rates=Uniform(30, 30), seed=seed)
preC = nengo.Ensemble(NPre, 1, max_rates=Uniform(30, 30), seed=seed)
fdfw = nengo.Ensemble(N, 1, neuron_type=Bio("Pyramidal", DA=DA), seed=seed)
ens = nengo.Ensemble(N, 1, neuron_type=Bio("Pyramidal", DA=DA), seed=seed+1)
inh = nengo.Ensemble(N, 1, neuron_type=Bio("Interneuron", DA=DA), seed=seed+2)
tarFdfw = nengo.Ensemble(N, 1, max_rates=Uniform(30, 30), intercepts=Uniform(-0.8, 0.8), neuron_type=nengo.LIF(), seed=seed)
tarEns = nengo.Ensemble(N, 1, max_rates=Uniform(30, 30), intercepts=Uniform(-0.8, 0.8), neuron_type=nengo.LIF(), seed=seed+1)
tarInh = nengo.Ensemble(N, 1, max_rates=Uniform(30, 30), intercepts=Uniform(0.2, 0.8), encoders=Choice([[1]]), neuron_type=nengo.LIF(), seed=seed+2)
cA = nengo.Connection(inptA, preA, synapse=None, seed=seed)
cB = nengo.Connection(inptB, preB, synapse=None, seed=seed)
cC = nengo.Connection(inptC, preC, synapse=None, seed=seed)
pInptA = nengo.Probe(inptA, synapse=None)
pInptB = nengo.Probe(inptB, synapse=None)
pInptC = nengo.Probe(inptC, synapse=None)
pPreA = nengo.Probe(preA.neurons, synapse=None)
pPreB = nengo.Probe(preB.neurons, synapse=None)
pPreC = nengo.Probe(preC.neurons, synapse=None)
pFdfw = nengo.Probe(fdfw.neurons, synapse=None)
pTarFdfw = nengo.Probe(tarFdfw.neurons, synapse=None)
pEns = nengo.Probe(ens.neurons, synapse=None)
pTarEns = nengo.Probe(tarEns.neurons, synapse=None)
pInh = nengo.Probe(inh.neurons, synapse=None)
pTarInh = nengo.Probe(tarInh.neurons, synapse=None)
if stage==1:
nengo.Connection(inptA, tarFdfw, synapse=fPre, seed=seed)
nengo.Connection(inptB, tarEns, synapse=fPre, seed=seed)
nengo.Connection(inptC, tarInh, synapse=fPre, seed=seed)
c1 = nengo.Connection(preA, fdfw, synapse=fPre, solver=NoSolver(dPreA), seed=seed)
c2 = nengo.Connection(preB, ens, synapse=fPre, solver=NoSolver(dPreB), seed=seed)
c3 = nengo.Connection(preC, inh, synapse=fPre, solver=NoSolver(dPreC), seed=seed)
learnEncoders(c1, tarFdfw, fS, alpha=alpha, eMax=eMax, tTrans=tTrans)
learnEncoders(c2, tarEns, fS, alpha=3*alpha, eMax=10*eMax, tTrans=tTrans)
learnEncoders(c3, tarInh, fS, alpha=alpha/3, eMax=eMax, tTrans=tTrans)
if stage==2:
c1 = nengo.Connection(preA, fdfw, synapse=fPre, solver=NoSolver(dPreA), seed=seed)
c2 = nengo.Connection(preC, inh, synapse=fPre, solver=NoSolver(dPreC), seed=seed)
if stage==3:
cB.synapse = fNMDA
nengo.Connection(inptA, tarFdfw, synapse=fPre, seed=seed)
nengo.Connection(inptB, tarEns, synapse=fPre, seed=seed)
nengo.Connection(tarFdfw, tarEns, synapse=fNMDA, transform=Tff, seed=seed)
c1 = nengo.Connection(preA, fdfw, synapse=fPre, solver=NoSolver(dPreA), seed=seed)
c2 = nengo.Connection(preB, ens, synapse=fPre, solver=NoSolver(dPreB), seed=seed)
c3 = nengo.Connection(fdfw, ens, synapse=NMDA(), solver=NoSolver(dFdfw), seed=seed)
learnEncoders(c3, tarEns, fS, alpha=alpha, eMax=eMax, tTrans=tTrans)
if stage==4:
cB.synapse = fNMDA
c1 = nengo.Connection(preA, fdfw, synapse=fPre, solver=NoSolver(dPreA), seed=seed)
c2 = nengo.Connection(preB, ens, synapse=fPre, solver=NoSolver(dPreB), seed=seed)
c3 = nengo.Connection(fdfw, ens, synapse=NMDA(), solver=NoSolver(dFdfw), seed=seed)
if stage==5:
preB2 = nengo.Ensemble(NPre, 1, max_rates=Uniform(30, 30), seed=seed)
ens2 = nengo.Ensemble(N, 1, neuron_type=Bio("Pyramidal", DA=DA), seed=seed+1)
ens3 = nengo.Ensemble(N, 1, neuron_type=Bio("Pyramidal", DA=DA), seed=seed+1)
nengo.Connection(inptB, preB2, synapse=fNMDA, seed=seed)
c1 = nengo.Connection(preA, fdfw, synapse=fPre, solver=NoSolver(dPreA), seed=seed)
c2 = nengo.Connection(fdfw, ens, synapse=NMDA(), solver=NoSolver(dFdfw), seed=seed)
c3 = nengo.Connection(preB, ens2, synapse=fPre, solver=NoSolver(dPreB), seed=seed)
c4 = nengo.Connection(ens2, ens, synapse=NMDA(), solver=NoSolver(dBio), seed=seed)
c5 = nengo.Connection(fdfw, ens3, synapse=NMDA(), solver=NoSolver(dFdfw), seed=seed)
c6 = nengo.Connection(preB2, ens3, synapse=fPre, solver=NoSolver(dPreB), seed=seed)
learnEncoders(c4, ens3, fS, alpha=alpha, eMax=eMax, tTrans=tTrans)
pTarEns = nengo.Probe(ens3.neurons, synapse=None)
if stage==6:
preA2 = nengo.Ensemble(NPre, 1, max_rates=Uniform(30, 30), seed=seed)
fdfw2 = nengo.Ensemble(N, 1, neuron_type=Bio("Pyramidal", DA=DA), seed=seed)
fdfw3 = nengo.Ensemble(N, 1, neuron_type=Bio("Pyramidal", DA=DA), seed=seed)
fdfw4 = nengo.Ensemble(N, 1, neuron_type=Bio("Pyramidal", DA=DA), seed=seed)
tarFdfw4 = nengo.Ensemble(N, 1, max_rates=Uniform(30, 30), intercepts=Uniform(-0.8, 0.8), neuron_type=nengo.LIF(), seed=seed)
nengo.Connection(inptA, tarFdfw4, synapse=fPre, seed=seed)
nengo.Connection(inptB, preA2, synapse=fNMDA, seed=seed)
nengo.Connection(inptC, tarInh, synapse=fPre, seed=seed)
nengo.Connection(tarInh, tarFdfw4.neurons, synapse=None, transform=-1e2*np.ones((N, 1)), seed=seed)
c1 = nengo.Connection(preB, ens, synapse=fPre, solver=NoSolver(dPreB), seed=seed)
c2 = nengo.Connection(ens, fdfw2, synapse=NMDA(), solver=NoSolver(dNeg), seed=seed)
c3 = nengo.Connection(preA2, fdfw3, synapse=fPre, solver=NoSolver(dPreA), seed=seed)
c4 = nengo.Connection(preA, fdfw4, synapse=fPre, solver=NoSolver(dPreA), seed=seed)
c5 = nengo.Connection(preC, inh, synapse=fPre, solver=NoSolver(dPreC), seed=seed)
c6 = nengo.Connection(inh, fdfw4, synapse=GABA(), solver=NoSolver(dInh), seed=seed)
learnEncoders(c2, fdfw3, fS, alpha=alpha, eMax=eMax, tTrans=tTrans)
learnEncoders(c6, tarFdfw4, fS, alpha=1e3*alpha, eMax=1e3*eMax, tTrans=tTrans, inh=True)
pFdfw2 = nengo.Probe(fdfw2.neurons, synapse=None)
pFdfw4 = nengo.Probe(fdfw4.neurons, synapse=None)
pTarFdfw2 = nengo.Probe(fdfw3.neurons, synapse=None)
pTarFdfw4 = nengo.Probe(tarFdfw4.neurons, synapse=None)
if stage==7:
c1 = nengo.Connection(preA, fdfw, synapse=fPre, solver=NoSolver(dPreA), seed=seed)
c2 = nengo.Connection(fdfw, ens, synapse=NMDA(), solver=NoSolver(dFdfw), seed=seed)
c3 = nengo.Connection(ens, ens, synapse=NMDA(), solver=NoSolver(dBio), seed=seed)
c4 = nengo.Connection(ens, fdfw, synapse=NMDA(), solver=NoSolver(dNeg), seed=seed)
c5 = nengo.Connection(preC, inh, synapse=fPre, solver=NoSolver(dPreC), seed=seed)
c6 = nengo.Connection(inh, fdfw, synapse=GABA(), solver=NoSolver(dInh), seed=seed)
with nengo.Simulator(model, seed=seed, dt=dt, progress_bar=False) as sim:
if stage==1:
setWeights(c1, dPreA, ePreA)
setWeights(c2, dPreB, ePreB)
setWeights(c3, dPreC, ePreC)
if stage==2:
setWeights(c1, dPreA, ePreA)
setWeights(c2, dPreC, ePreC)
if stage==3:
setWeights(c1, dPreA, ePreA)
setWeights(c2, dPreB, ePreB)
setWeights(c3, dFdfw, eFdfw)
if stage==4:
setWeights(c1, dPreA, ePreA)
setWeights(c2, dPreB, ePreB)
setWeights(c3, dFdfw, eFdfw)
if stage==5:
setWeights(c1, dPreA, ePreA)
setWeights(c2, dFdfw, eFdfw)
setWeights(c3, dPreB, ePreB)
setWeights(c4, dBio, eBio)
setWeights(c5, dFdfw, eFdfw)
setWeights(c6, dPreB, ePreB)
if stage==6:
setWeights(c1, dPreB, ePreB)
setWeights(c2, dNeg, eNeg)
setWeights(c3, dPreA, ePreA)
setWeights(c4, dPreA, ePreA)
setWeights(c5, dPreC, ePreC)
setWeights(c6, dInh, eInh)
if stage==7:
setWeights(c1, dPreA, ePreA)
setWeights(c2, dFdfw, eFdfw)
setWeights(c3, dBio, eBio)
setWeights(c4, dNeg, eNeg)
setWeights(c5, dPreC, ePreC)
setWeights(c6, dInh, eInh)
neuron.h.init()
sim.run(t, progress_bar=True)
reset_neuron(sim, model)
ePreA = c1.e if stage==1 else ePreA
ePreB = c2.e if stage==1 else ePreB
ePreC = c3.e if stage==1 else ePreC
eFdfw = c3.e if stage==3 else eFdfw
eBio = c4.e if stage==5 else eBio
eNeg = c2.e if stage==6 else eNeg
eInh = c6.e if stage==6 else eInh
return dict(
times=sim.trange(),
inptA=sim.data[pInptA],
inptB=sim.data[pInptB],
inptC=sim.data[pInptC],
preA=sim.data[pPreA],
preB=sim.data[pPreB],
preC=sim.data[pPreC],
fdfw=sim.data[pFdfw],
ens=sim.data[pEns],
inh=sim.data[pInh],
tarFdfw=sim.data[pTarFdfw],
tarEns=sim.data[pTarEns],
tarInh=sim.data[pTarInh],
fdfw2=sim.data[pFdfw2] if stage==6 else None,
fdfw4=sim.data[pFdfw4] if stage==6 else None,
tarFdfw2=sim.data[pTarFdfw2] if stage==6 else None,
tarFdfw4=sim.data[pTarFdfw4] if stage==6 else None,
ePreA=ePreA,
ePreB=ePreB,
ePreC=ePreC,
eFdfw=eFdfw,
eBio=eBio,
eNeg=eNeg,
eInh=eInh,
)
f"\n\tEncoders: {pre.encoders}\n\tIntercepts: {pre.intercepts}\n\tMax rates: {pre.max_rates}" )
f.write(
f"Post:\n\t {post.neuron_type} \n\tNeurons: {post.n_neurons}\n\tGain: {post.gain} \n\tBias: {post.bias} "
f"\n\tEncoders: {post.encoders}\n\tIntercepts: {post.intercepts}\n\tMax rates: {post.max_rates}" )
f.write( f"Rule:\n\t {conn.learning_rule_type}\n" )
if args.level >= 1:
print( "######################################################",
"####################### TRAINING #####################",
"######################################################",
sep="\n" )
print( f"Backend is {args.backend}, running on ", end="" )
if args.backend == "nengo_core":
print( "CPU" )
with nengo.Simulator( model, seed=args.seed ) as sim_train:
sim_train.run( sim_train_time )
if args.backend == "nengo_dl":
print( args.device )
with nengo_dl.Simulator( model, seed=args.seed, device=args.device ) as sim_train:
sim_train.run( sim_train_time )
# print number of recorded spikes
num_spikes_train = np.sum( sim_train.data[ post_probe ] > 0, axis=0 )
for line in graph.graph( f"Spikes distribution (timestep={dt}):",
[ (str( i ), x / np.sum( num_spikes_train ) * 100) for i, x in
enumerate( num_spikes_train ) ] ):
print( line )
print( "\tTotal:", np.sum( num_spikes_train ) )
print( f"\tNormalised standard dev.: {np.std( num_spikes_train ) / np.mean( num_spikes_train )}" )
nengo.Connection(dummy_ctx, cortex1, function=project)
# prior is sent to the node 'cortex_in' to be processed
# and converted to the posterior.
nengo.Connection(cortex1, cortex_in)
### ------------------------------------------------------------------------------------------------------- ###
# Probes
# wta
wta_doutp = nengo.Probe(wta.output, synapse=0.02)
# ctx
cortex1_p = nengo.Probe(cortex1, synapse=0.02)
sim = nengo.Simulator(model, dt=dt) # Create the simulator
sim.run(training_size * 3) # Run it for 1 second
# Collect data and write it to a pickle file
data_out = {}
data_out[index] = [
sim.data[wta_doutp], sim.data[cortex1_p], space._basis, space._scale
]
pickleout = open("data_out/" + fname_output + "_" + str(index) + ".p",
'wb')
pickle.dump(data_out, pickleout)
pickleout.close()
print("pickle complete for iteration: ", index, "in ", fname_output)
# ------------------------------------------------------------------------------------------------------- #
def run_experiment(sacc, t):
sim = nengo.Simulator(model, dt = 0.001, builder=nengo.builder.Builder(copy=False))
sim.run(t)
return sim.data(MN_hr_p), sim.data(MN_vr_p)
def evaluate(self, p, plt):
t_tone_start = 0.0
t_tone_end = t_tone_start + p.t_tone
t_puff_start = t_tone_end + p.t_delay
t_puff_end = t_puff_start + p.t_puff
def puff_func(t):
if t_puff_start < t % p.period < t_puff_end:
return 1
else:
return 0
def tone_func(t):
if t_tone_start < t % p.period < t_tone_end:
return 1
else:
return 0
model = nengo.Network()
with model:
###########################################################################
# Setup the conditioned stimulus (i.e., a tone) and the unconditioned #
# stimulus (i.e., a puff) #
###########################################################################
nd_tone = nengo.Node(tone_func)
nd_puff = nengo.Node(puff_func)
###########################################################################
# Setup the reflex generator and the eye-motor system #
###########################################################################
# The reflex pathway is across the Trigeminal nucleus in the brainstem;
# we don't model this in this particular model
# Scaling factor that has to be applied to the reflex trajectory to scale
# it to a range from 0 to 1
reflex_scale = 1.0 / 25.0
# The reflex system takes an input and produces the reflex trajectory on
# the rising edge (convolves the differential of the input with the
# trajectory)
nd_reflex = nengo.Node(EyeblinkReflex()) # Unscaled output
nd_reflex_out = nengo.Node(size_in=1) # Normalised output
nengo.Connection(nd_reflex[0],
nd_reflex_out,
transform=reflex_scale,
synapse=None)
if not p.do_minimal:
# The eyelid component represents the state of the eye in the world.
# It receives two inputs, an agonist input (closing the eye, dim 0) and an
# antagonist input (opening the eye, dim 1).
nd_eyelid = nengo.Node(Eyelid()) # Unscaled input
eyelid_in = nengo.Ensemble(n_neurons=100, dimensions=2)
nengo.Connection(eyelid_in,
nd_eyelid[0],
transform=1.0 / reflex_scale,
function=lambda x: max(x[0], x[1]),
synapse=0.005)
# Constantly open the eye a little bit
nd_eye_bias = nengo.Node(p.eye_bias)
nengo.Connection(nd_eye_bias, nd_eyelid[1])
# We can't detect the puff if the eye is closed, multiply the output from
# nd_puff with the amount the eye is opened. This is our unconditioned
# stimulus
# NOTE: Currently disabled by commenting out the line below
c0, c1 = nengo.Node(size_in=1), nengo.Node(
size_in=1) # Only for GUI
nengo.Connection(nd_eyelid, c0, synapse=None)
nengo.Connection(c0, c1, synapse=None)
# nengo.Connection(c1, nd_us[1], synapse=None)
# Connect the unconditioned stimulus to the reflex generator
nd_us = nengo.Node(lambda t, x: x[0] * (1 - x[1]),
size_in=2,
size_out=1)
nengo.Connection(nd_puff, nd_us[0], synapse=None)
nengo.Connection(nd_us, nd_reflex)
if not p.do_minimal:
nengo.Connection(nd_reflex_out, eyelid_in[0])
###########################################################################
# Generate a neural representation of the conditioned stimulus #
###########################################################################
nd_cs = nengo.Node(size_in=1)
ens_pcn = nengo.Ensemble(n_neurons=100, dimensions=1)
nengo.Connection(nd_tone, nd_cs, synapse=None)
nengo.Connection(nd_cs, ens_pcn)
###########################################################################
# Generate a LMU representation of the conditioned stimulus #
###########################################################################
# Build the LMU, feed the conditioned stimulus into it
net_granule_golgi = GranuleGolgiCircuit(
ens_pcn,
tau=p.tau,
q=p.q,
theta=p.theta,
n_granule=p.n_granule,
n_golgi=p.n_granule // 10,
golgi_intercepts=nengo.dists.CosineSimilarity(p.q + 2)
if p.use_cosine else nengo.dists.Uniform(-1, 1),
granule_intercepts=nengo.dists.CosineSimilarity(p.q + 2)
if p.use_cosine else nengo.dists.Uniform(-1, 1),
mode=p.mode,
)
###########################################################################
# Learn the connection from the Granule cells to the Purkinje cells via #
# input from the Interior Olive #
###########################################################################
# This is the US pathway; the data is relayed from the Trigeminal nucleus
# to the Interior Olive.
ens_cn = nengo.Ensemble(n_neurons=100, dimensions=1)
ens_io = nengo.Ensemble(n_neurons=100, dimensions=1)
ens_purkinje = nengo.Ensemble(n_neurons=100, dimensions=1)
# Represent the error signal in ens_io
nengo.Connection(nd_reflex_out[0], ens_io, transform=-1)
nengo.Connection(
ens_cn, ens_io, transform=1,
synapse=p.tau_error) # This connection does not exist
# Project from the
c_learn = nengo.Connection(
net_granule_golgi.ens_granule.neurons,
ens_purkinje,
transform=np.zeros((ens_purkinje.dimensions,
net_granule_golgi.ens_granule.n_neurons)),
learning_rule_type=nengo.learning_rules.PES(
learning_rate=p.learning_rate, pre_synapse=p.tau_pre))
nengo.Connection(ens_io, c_learn.learning_rule)
###########################################################################
# Project from CN onto the motor system
###########################################################################
nengo.Connection(ens_purkinje, ens_cn)
if not p.do_minimal:
nengo.Connection(ens_cn, eyelid_in[1])
p_nd_reflex_out = nengo.Probe(nd_reflex_out,
sample_every=p.sample_every)
if not p.do_minimal:
p_eyelid = nengo.Probe(nd_eyelid, sample_every=p.sample_every)
p_purkinje = nengo.Probe(ens_purkinje,
synapse=p.tau_purkinje,
sample_every=p.sample_every)
p_granule = nengo.Probe(net_granule_golgi.ens_granule,
synapse=0.03,
sample_every=p.sample_every)
add_labels(model, locals=locals())
sim = nengo.Simulator(model)
with sim:
sim.run(p.period * p.n_trials)
dt = p.sample_every
steps = int(p.period / dt)
purk = sim.data[p_purkinje].reshape(-1, steps).T
v = np.clip(purk[:, :], 0, np.inf) * dt / reflex_scale
pos = np.cumsum(v, axis=0)
if plt:
t = np.arange(steps) * dt
ax1 = plt.subplot(4, 1, 1)
ax1.set_ylabel('granule')
ax2 = plt.subplot(4, 1, 2)
ax2.set_ylabel('purkinje')
ax3 = plt.subplot(4, 1, 3)
ax3.set_ylabel('eye position\n(due to reflex)')
ax4 = plt.subplot(4, 1, 4)
ax4.set_ylabel('eye position\n at puff start')
ax4.set_xlabel('trial')
n_steps = len(sim.data[p_purkinje])
cmap = matplotlib.cm.get_cmap("viridis")
for i in range(0, n_steps, steps):
color = cmap(i / n_steps)
#if not p.do_minimal:
# ax1.plot(t, sim.data[p_eyelid][i:i+steps], label='eyelid %d'%(i//steps), ls='--')
ax2.plot(t, sim.data[p_purkinje][i:i + steps], color=color)
ax3.plot(t,
np.cumsum(np.abs(sim.data[p_purkinje][i:i + steps])) *
dt / reflex_scale,
color=color)
ax1.plot(t, sim.data[p_granule][:steps])
ax2b = ax2.twinx()
ax2b.plot(t, sim.data[p_nd_reflex_out][:steps], c='k', ls='--')
ax4.plot(pos[int(t_puff_start / dt)])
r = dict(
final_pos=pos[int(t_puff_start / dt), -1],
pos_at_puff_start=pos[int(t_puff_start / dt)],
)
if p.save_plot_data:
r['purkinje'] = sim.data[p_purkinje]
r['granule'] = sim.data[p_granule][:steps]
r['reflex'] = sim.data[p_nd_reflex_out][:steps]
return r
p_exc_rates = nengo.Probe(srf_network.exc.neurons)
p_inh_rates = nengo.Probe(srf_network.inh.neurons)
p_inh_weights = nengo.Probe(srf_network.conn_I, 'weights', sample_every=1.0)
p_exc_weights = nengo.Probe(srf_network.conn_E, 'weights', sample_every=1.0)
if srf_network.conn_EE is not None:
p_rec_weights = nengo.Probe(srf_network.conn_EE, 'weights', sample_every=1.0)
p_weights_PV_E = nengo.Probe(conn_PV_E, 'weights', sample_every=1.0)
p_weights_PV_I = nengo.Probe(conn_PV_I, 'weights', sample_every=1.0)
p_weights_decoder_I = nengo.Probe(conn_decoder_I, 'weights', sample_every=1.0)
p_decoder_spikes = nengo.Probe(decoder.neurons, synapse=None)
p_decoder_inh_rates = nengo.Probe(decoder_inh.neurons, synapse=None)
p_cd = nengo.Probe(coincidence_detection, synapse=None, sample_every=0.1)
with nengo.Simulator(model, optimize=True, progress_bar=TerminalProgressBar()) as sim:
sim.run(np.max(t_end))
output_spikes = sim.data[p_output_spikes]
exc_rates = sim.data[p_exc_rates]
inh_rates = sim.data[p_inh_rates]
decoder_spikes = sim.data[p_decoder_spikes]
decoder_inh_rates = sim.data[p_decoder_inh_rates]
output_rates = rates_kernel(sim.trange(), output_spikes, tau=0.1)
decoder_rates = rates_kernel(sim.trange(), decoder_spikes, tau=0.1)
exc_weights = sim.data[p_exc_weights]
#inh_weights = sim.data[p_inh_weights]
np.save("output_spikes_dof.npy", output_spikes)
np.save("output_rates_dof.npy", output_rates)
def test_conv_connection(channels, channels_last, Simulator, seed, rng, plt,
allclose):
if channels_last:
plt.saveas = None
pytest.xfail("Blocked by CxBase cannot be > 256 bug")
# load data
with open(os.path.join(test_dir, 'mnist10.pkl'), 'rb') as f:
test10 = pickle.load(f)
test_x = test10[0][0].reshape(28, 28)
test_x = 1.999 * test_x - 0.999 # range (-1, 1)
input_shape = nengo_transforms.ChannelShape(
(test_x.shape + (channels, )) if channels_last else
((channels, ) + test_x.shape),
channels_last=channels_last)
filters = Gabor(freq=Uniform(0.5, 1)).generate(8, (7, 7), rng=rng)
filters = filters[None, :, :, :] # single channel
filters = np.transpose(filters, (2, 3, 0, 1))
strides = (2, 2)
tau_rc = 0.02
tau_ref = 0.002
tau_s = 0.005
dt = 0.001
neuron_type = LoihiLIF(tau_rc=tau_rc, tau_ref=tau_ref)
pres_time = 0.1
with nengo.Network(seed=seed) as model:
nengo_loihi.add_params(model)
u = nengo.Node(test_x.ravel(), label="u")
a = nengo.Ensemble(input_shape.size,
1,
neuron_type=LoihiSpikingRectifiedLinear(),
max_rates=nengo.dists.Choice([40 / channels]),
intercepts=nengo.dists.Choice([0]),
label='a')
model.config[a].on_chip = False
if channels == 1:
nengo.Connection(u, a.neurons, transform=1, synapse=None)
elif channels == 2:
# encode image into spikes using two channels (on/off)
if input_shape.channels_last:
nengo.Connection(u, a.neurons[0::2], transform=1, synapse=None)
nengo.Connection(u,
a.neurons[1::2],
transform=-1,
synapse=None)
else:
k = input_shape.spatial_shape[0] * input_shape.spatial_shape[1]
nengo.Connection(u, a.neurons[:k], transform=1, synapse=None)
nengo.Connection(u, a.neurons[k:], transform=-1, synapse=None)
filters = np.concatenate([filters, -filters], axis=2)
else:
raise ValueError("Test not configured for more than two channels")
conv2d_transform = nengo_transforms.Convolution(
8,
input_shape,
strides=strides,
kernel_size=(7, 7),
channels_last=channels_last,
init=filters)
output_shape = conv2d_transform.output_shape
gain, bias = neuron_type.gain_bias(max_rates=100, intercepts=0)
gain = gain * 0.01 # account for `a` max_rates
b = nengo.Ensemble(output_shape.size,
1,
neuron_type=neuron_type,
gain=nengo.dists.Choice([gain[0]]),
bias=nengo.dists.Choice([bias[0]]),
label='b')
nengo.Connection(a.neurons,
b.neurons,
synapse=tau_s,
transform=conv2d_transform)
bp = nengo.Probe(b.neurons)
with nengo.Simulator(model, dt=dt, optimize=False) as sim:
sim.run(pres_time)
ref_out = sim.data[bp].mean(axis=0).reshape(output_shape.shape)
# Currently, non-gpu TensorFlow does not support channels first in conv
use_nengo_dl = HAS_DL and channels_last
ndl_out = np.zeros_like(ref_out)
if use_nengo_dl:
with nengo_dl.Simulator(model, dt=dt) as sim_dl:
sim_dl.run(pres_time)
ndl_out = sim_dl.data[bp].mean(axis=0).reshape(output_shape.shape)
with nengo_loihi.Simulator(model, dt=dt, target='simreal') as sim_real:
sim_real.run(pres_time)
real_out = sim_real.data[bp].mean(axis=0).reshape(output_shape.shape)
with Simulator(model, dt=dt) as sim_loihi:
if "loihi" in sim_loihi.sims:
sim_loihi.sims["loihi"].snip_max_spikes_per_step = 800
sim_loihi.run(pres_time)
sim_out = sim_loihi.data[bp].mean(axis=0).reshape(output_shape.shape)
if not output_shape.channels_last:
ref_out = np.transpose(ref_out, (1, 2, 0))
ndl_out = np.transpose(ndl_out, (1, 2, 0))
real_out = np.transpose(real_out, (1, 2, 0))
sim_out = np.transpose(sim_out, (1, 2, 0))
out_max = max(ref_out.max(), sim_out.max())
# --- plot results
rows = 2
cols = 3
ax = plt.subplot(rows, cols, 1)
imshow(test_x, vmin=0, vmax=1, ax=ax)
ax = plt.subplot(rows, cols, 2)
tile(np.transpose(filters[0], (2, 0, 1)), cols=8, ax=ax)
ax = plt.subplot(rows, cols, 3)
plt.hist(ref_out.ravel(), bins=31)
plt.hist(sim_out.ravel(), bins=31)
ax = plt.subplot(rows, cols, 4)
tile(np.transpose(ref_out, (2, 0, 1)), vmin=0, vmax=out_max, cols=8, ax=ax)
ax = plt.subplot(rows, cols, 5)
tile(np.transpose(ndl_out, (2, 0, 1)), vmin=0, vmax=out_max, cols=8, ax=ax)
ax = plt.subplot(rows, cols, 6)
tile(np.transpose(sim_out, (2, 0, 1)), vmin=0, vmax=out_max, cols=8, ax=ax)
if use_nengo_dl:
assert allclose(ndl_out, ref_out, atol=1e-5, rtol=1e-5)
assert allclose(real_out, ref_out, atol=1, rtol=1e-3)
assert allclose(sim_out, ref_out, atol=10, rtol=1e-3)
def __init__(self,
mapping,
threshold=0.2,
learning_rate=1e-4,
DimVocab=256,
subdimensions=32,
learned_function_scale=2.0):
model = spa.SPA()
self.model = model
self.mapping = mapping
#self.vocab_category = spa.Vocabulary(D_category)
#self.vocab_items = spa.Vocabulary(D_items)
self.VocabUnified = spa.Vocabulary(DimVocab)
for k in sorted(
mapping.keys()): #allocating verctors for categories name
self.VocabUnified.parse(k)
for v in mapping[k]: # allocating vectors to the items
self.VocabUnified.parse(v)
with model:
model.cue = spa.State(DimVocab,
vocab=self.VocabUnified,
subdimensions=subdimensions)
model.target = spa.State(DimVocab,
vocab=self.VocabUnified,
subdimensions=subdimensions)
c = []
n_sub = len(model.cue.state_ensembles.ea_ensembles)
for i in range(n_sub):
cues = []
targets = []
for cue, vals in mapping.items():
for val in vals:
cues.append(
self.VocabUnified.parse(cue).v[i *
subdimensions:(i +
1) *
subdimensions])
targets.append(
self.VocabUnified.parse(val).v / n_sub *
learned_function_scale)
cues.append(
self.VocabUnified.parse(val).v[i *
subdimensions:(i +
1) *
subdimensions])
targets.append(
self.VocabUnified.parse(cue).v / n_sub *
learned_function_scale)
cc = nengo.Connection(
model.cue.all_ensembles[i],
model.target.input,
learning_rule_type=nengo.PES(learning_rate=learning_rate),
**nengo.utils.connection.target_function(cues, targets))
cc.eval_points = cues
cc.function = targets
c.append(cc)
print i
#model.error = spa.State(D_items, vocab=self.vocab_items)
#nengo.Connection(model.items.output, model.error.input)
#nengo.Connection(model.error.output, c.learning_rule)
#I am not sure how this implements the learning
model.error = spa.State(DimVocab,
vocab=self.VocabUnified,
subdimensions=subdimensions)
nengo.Connection(model.target.output, model.error.input)
print 'the loop ended, right?'
for cc in c:
nengo.Connection(model.error.output, cc.learning_rule)
self.stim_cue_value = np.zeros(DimVocab) #?
self.stim_cue = nengo.Node(self.stim_cue_fun) #?
nengo.Connection(self.stim_cue, model.cue.input, synapse=None)
self.stim_correct_value = np.zeros(DimVocab)
self.stim_correct = nengo.Node(self.stim_correct)
nengo.Connection(self.stim_correct,
model.error.input,
synapse=None,
transform=-1)
self.stim_stoplearn_value = np.zeros(1)
self.stim_stoplearn = nengo.Node(self.stim_stoplearn)
for ens in model.error.all_ensembles:
nengo.Connection(self.stim_stoplearn,
ens.neurons,
synapse=None,
transform=-10 * np.ones((ens.n_neurons, 1)))
self.stim_justmemorize_value = np.zeros(1)
self.stim_justmemorize = nengo.Node(self.stim_justmemorize)
for ens in model.target.all_ensembles: #?
nengo.Connection(self.stim_justmemorize,
ens.neurons,
synapse=None,
transform=-10 * np.ones((ens.n_neurons, 1)))
self.probe_target = nengo.Probe(model.target.output, synapse=0.01)
if __name__ != '__builtin__':
self.sim = nengo.Simulator(self.model)
def test_conv_input(channels_last, Simulator, plt, allclose):
input_shape = nengo_transforms.ChannelShape((4, 4, 1) if channels_last else
(1, 4, 4),
channels_last=channels_last)
seed = 3 # fix seed to do the same computation for both channel positions
rng = np.random.RandomState(seed + 1)
with nengo.Network(seed=seed) as net:
nengo_loihi.add_params(net)
a = nengo.Node(rng.uniform(0, 1, size=input_shape.size))
nc = 2
kernel = np.array([1., -1.]).reshape((1, 1, 1, nc))
transform = nengo_transforms.Convolution(nc,
input_shape,
channels_last=channels_last,
init=kernel,
kernel_size=(1, 1))
b = nengo.Ensemble(transform.output_shape.size,
1,
neuron_type=nengo.SpikingRectifiedLinear(),
max_rates=nengo.dists.Choice([50]),
intercepts=nengo.dists.Choice([0]))
net.config[b].on_chip = False
nengo.Connection(a, b.neurons, transform=transform)
output_shape = transform.output_shape
nf = 4
kernel = rng.uniform(-0.005, 0.005, size=(3, 3, nc, nf))
transform = nengo_transforms.Convolution(nf,
output_shape,
channels_last=channels_last,
init=-kernel,
kernel_size=(3, 3))
c = nengo.Ensemble(transform.output_shape.size,
1,
neuron_type=nengo.LIF(),
max_rates=nengo.dists.Choice([100]),
intercepts=nengo.dists.Choice([0]))
nengo.Connection(b.neurons, c.neurons, transform=transform)
output_shape = transform.output_shape
p = nengo.Probe(c.neurons)
with nengo.Simulator(net, optimize=False) as sim:
sim.run(1.0)
with Simulator(net, seed=seed) as sim_loihi:
sim_loihi.run(1.0)
p0 = np.sum(sim.data[p] > 0, axis=0).reshape(output_shape.shape)
p1 = np.sum(sim_loihi.data[p] > 0, axis=0).reshape(output_shape.shape)
if not channels_last:
p0 = np.transpose(p0, (1, 2, 0))
p1 = np.transpose(p1, (1, 2, 0))
plt.plot(p0.ravel(), 'k')
plt.plot(p1.ravel(), 'b--')
# loihi spikes are not exactly the same, but should be close-ish
assert allclose(p0, p1, rtol=0.15, atol=1)
def run_model(self):
res = ProductionResult()
rng = np.random.RandomState(self.seed)
dt = self.model.trial.dt
paths, freqs = analysis.get_syllables(self.n_syllables, self.minfreq,
self.maxfreq, rng)
t = 0.8 # 0.2 for initial stuff, 0.6 fudge factor
gs_targets = []
for path, freq in zip(paths, freqs):
gs = vtl.parse_ges(path)
gs_targets.append(gs)
self.model.add_syllable(label=path2label(path),
freq=freq,
trajectory=gs.trajectory(dt))
t += 1. / freq
# Determine and save syllable sequence
if self.model.trial.repeat:
seq_ix = rng.randint(len(paths), size=self.sequence_len)
else:
seq_ix = rng.permutation(len(paths))[:self.sequence_len]
seq = [path2label(paths[i]) for i in seq_ix]
seq_str = " + ".join(
["%s*POS%d" % (label, i + 1) for i, label in enumerate(seq)])
self.model.trial.sequence = seq_str
res.seq = np.array(seq)
# Use a minimum number of syllables
self.model.sequence.n_positions = self.n_syllables
# Save frequencies for that sequence
res.freqs = np.array([self.model.syllables[i].freq for i in seq_ix])
net = self.model.build(nengo.Network(seed=self.seed))
with net:
p_out = nengo.Probe(net.production_info.output, synapse=0.01)
sim = nengo.Simulator(net)
sim.run(t)
# Get ideal trajectory; compare RMSE
delay_frames = int(self.model.trial.t_release / dt)
traj = ideal_traj(self.model, seq)
res.traj = traj
simtraj = sim.data[p_out][delay_frames:]
simtraj = simtraj[:traj.shape[0]]
traj = traj[:simtraj.shape[0]]
res.simtraj = simtraj
res.simrmse = npext.rmse(traj, simtraj)
# Reconstruct gesture score; compare to originals
gs = gesture_score(simtraj, dt)
res.accuracy, res.n_sub, res.n_del, res.n_ins = (analysis.gs_accuracy(
gs, gs_targets))
res.timing_mean, res.timing_var = analysis.gs_timing(gs, gs_targets)
res.cooccur, res.co_chance = analysis.gs_cooccur(gs, gs_targets)
log("Accuracy: %.3f" % res.accuracy)
# Get the reconstructed trajectory and audio
reconstructed = gs.trajectory(dt=dt)
res.reconstructed = reconstructed
minsize = min(reconstructed.shape[0], traj.shape[0])
res.reconstructedrmse = npext.rmse(traj[:minsize],
reconstructed[:minsize])
audio, fs = gs.synthesize()
res.audio = audio
res.fs = fs
tgt_gs = analysis.gs_combine([vtl.parse_ges(paths[i]) for i in seq_ix])
audio, _ = tgt_gs.synthesize()
res.clean_audio = audio
return res
def run_trial(params,
ens_seed=1337,
train_trials_seed=1337,
test_trials_seed=1234,
train_trials_per_cond=5,
test_trials_per_cond=5):
# Create model for training
srate = 1000
for key in ['q', 'n_neurons']:
params[key] = int(params[key])
train_model, train_probes = make_lmu_dms(
theta=params['theta'],
q=params['q'],
n_neurons=params['n_neurons'],
seed=ens_seed,
out_transform=None,
n_trials_per_cond=train_trials_per_cond,
trial_seed=train_trials_seed,
tau=params['tau'])
# Generate training data
train_sim = nengo.Simulator(train_model)
n_train_trials = train_trials_per_cond * 8 * 2 # 16 conditions
train_sim.run(6 * n_train_trials) # 6 seconds per trial
train_sim.close()
# Solve for weights - Data are generally too large for nengo's filt so we slice
# them to only evaluate specific ranges.
eval_ranges = [(0, 0.25), (0.75, 1.75), (2.75, 3.75), (4.75, 6.0)]
trial_tvec = np.arange(0, 6.0, 1 / srate)
b_eval = np.zeros(trial_tvec.shape, dtype=bool)
for t_win in eval_ranges:
b_eval = np.logical_or(
b_eval,
np.logical_and(t_win[0] <= trial_tvec, trial_tvec < t_win[1]))
b_eval = np.tile(b_eval, n_train_trials)
filt = nengo.synapses.Lowpass(0.01)
Y = filt.filt(train_sim.data[train_probes['ideal']][b_eval])
A = filt.filt(train_sim.data[train_probes['ensemble']][b_eval])
D, info = nengo.solvers.LstsqL2(solver=lss.LSMRScipy())(A, Y)
Y = A = None
# Create a new model with the learned weights
test_model, test_probes = make_lmu_dms(
theta=params['theta'],
q=params['q'],
n_neurons=params['n_neurons'],
seed=ens_seed,
out_transform=D.T,
n_trials_per_cond=test_trials_per_cond,
trial_seed=test_trials_seed,
tau=params['tau'])
# Run test simulation - break after N trials to report cumulative accuracy
def get_labels_from_sim(sim, n_trials):
samps = 6 * srate * n_trials
Y = sim.data[test_probes['ideal']][-samps:]
A = sim.data[test_probes['output']][-samps:]
b_slice = Y != 0
label = Y[b_slice].reshape(n_trials, -1)[:, 0]
score = np.mean(A[b_slice].reshape(n_trials, -1), axis=-1)
return label, score
total_test_trials = test_trials_per_cond * 8 * 2
test_sim = nengo.Simulator(test_model)
test_ix = min(20, total_test_trials)
test_sim.run(6 * test_ix)
labels, scores = get_labels_from_sim(test_sim, test_ix)
trial_step = 10
while test_ix < total_test_trials:
test_sim.run(6 * min(trial_step, total_test_trials - test_ix))
label, score = get_labels_from_sim(test_sim, trial_step)
labels = np.append(labels, label)
scores = np.append(scores, score)
fpr, tpr, thresholds = metrics.roc_curve(labels, scores)
test_auc = metrics.auc(fpr, tpr)
nni.report_intermediate_result(test_auc)
test_ix += trial_step
test_sim.close()
logger.debug('test_sim.close() returned.')
logger.debug('Final result is: %d', test_auc)
nni.report_final_result(test_auc)
logger.debug('Sent final result. Trial complete.')
def run_model(self):
res = RecognitionResult()
rng = np.random.RandomState(self.seed)
paths, freqs = analysis.get_syllables(self.n_syllables, self.minfreq,
self.maxfreq, rng)
for path, freq in zip(paths, freqs):
traj = vtl.parse_ges(path).trajectory(self.model.trial.dt)
self.model.add_syllable(label=path2label(path),
freq=freq,
trajectory=traj)
# Determine and save syllable sequence
if self.model.trial.repeat:
seq_ix = rng.randint(len(paths), size=self.sequence_len)
else:
seq_ix = rng.permutation(len(paths))[:self.sequence_len]
seq = [path2label(paths[i]) for i in seq_ix]
res.seq = np.array(seq)
# Determine how long to run
simt = 0.0
tgt_time = []
for label in res.seq:
syllable = self.model.syllable_dict[label]
simt += 1. / syllable.freq
tgt_time.append(simt)
# Set that sequence in the model
traj = ideal_traj(self.model, seq)
self.model.trial.trajectory = traj
# Save frequencies for that sequence
res.freqs = np.array([self.model.syllables[i].freq for i in seq_ix])
# -- Run the model
net = self.model.build(nengo.Network(seed=self.seed))
with net:
p_dmps = [
nengo.Probe(dmp.state[0], synapse=0.01)
for dmp in net.syllables
]
p_class = nengo.Probe(net.classifier, synapse=0.01)
p_mem = nengo.Probe(net.memory.output, synapse=0.01)
sim = nengo.Simulator(net)
sim.run(simt)
# Save iDMP system states
res.dmps = np.hstack([sim.data[p_d] for p_d in p_dmps])
res.dmp_labels = np.array([s.label for s in self.model.syllables])
# Save working memory similarities
res.memory = spa.similarity(sim.data[p_mem], net.vocab, True)
# Determine classification times and labels
t_ix, class_ix = analysis.classinfo(sim.data[p_class], res.dmps)
res.class_time = sim.trange()[t_ix]
res.class_labels = np.array([path2label(paths[ix]) for ix in class_ix])
# Calculate accuracy / timing metrics
recinfo = [(t, l) for t, l in zip(res.class_time, res.class_labels)]
tgtinfo = [(t, l) for t, l in zip(tgt_time, res.seq)]
res.acc, res.n_sub, res.n_del, res.n_ins = (analysis.cl_accuracy(
recinfo, tgtinfo))
res.tdiff_mean, res.tdiff_var = analysis.cl_timing(recinfo, tgtinfo)
log("Accuracy: %.3f" % res.acc)
# Determine if memory representation is correct
tgt_time = np.asarray(tgt_time)
mem_times = (tgt_time[1:] + tgt_time[:-1]) * 0.5
mem_ix = (mem_times / self.model.trial.dt).astype(int)
mem_class = np.argmax(res.memory[mem_ix], axis=1)
slabels = [s.label for s in self.model.syllables]
actual = np.array([slabels.index(lbl) for lbl in res.seq[:-1]])
res.memory_acc = np.mean(mem_class == actual)
return res
if run_in_GUI:
# to run in GUI, comment out next 4 lines for running without GUI
import nengo_gui
nengo_gui.GUI(model=model, filename=__file__, locals=locals(),
interactive=False, allow_file_change=False).start()
import sys
sys.exit()
else:
# to run in command line
with model:
probe_input = nengo.Probe(input_node)
probe_arm = nengo.Probe(arm_node[arm.DOF*2])
print 'building model...'
sim = nengo.Simulator(model, dt=.001)
print 'build complete.'
sim.run(10)
t = sim.trange()
x = sim.data[probe_arm]
y = sim.data[probe_arm]
# plot collected data
import matplotlib.pyplot as plt
plt.subplot(311)
plt.plot(t, x)
plt.xlabel('time')
plt.ylabel('probe_arm0')
def __init__(
self,
n_input, # Number of inputs (angles, velocities)
n_output, # Number of outputs (forces for joints)
n_neurons=1000, # Number of neurons / ensemble
n_ensembles=1, # Number of ensembles
seed=None, # Seed for random number generation
pes_learning_rate=1e-6, # Adaptation learning rate
means=None, # Means and variances to scale data from -1 to 1
variances=None, # Outliers will be scaled outside the -1 to 1
**kwargs):
self.n_neurons = n_neurons
self.n_ensembles = n_ensembles
self.pes_learning_rate = pes_learning_rate
self.input_signal = np.zeros(n_input)
self.training_signal = np.zeros(n_output)
self.output = np.zeros(n_output)
np.random.seed = 42 #seed
# Accounting for spherical hyperspace
n_input += 1
# Accounting for unknow means and variances ---------------------------------
if means is not None and variances is None:
variances = np.ones(means.shape)
elif means is None and variances is not None:
means = np.zeros(variances.shape)
self.means = np.asarray(means)
self.variances = np.asarray(variances)
# Setting synapse time constants --------------------------------------------
self.tau_input = 0.012 # Time constant for input connection
self.tau_training = 0.012 # Time constant for training signal
self.tau_output = 0.2 # Time constant for the output
# Setting intercepts for the ensembles --------------------------------------
# Generates intercepts for a d-dimensional ensemble, such that, given a
# random uniform input (from the interior of the d-dimensional ball), the
# probability of a neuron firing has the probability density function given
# by rng.triangular(left, mode, right, size=n)
triangular = np.random.triangular(left=0.35,
mode=0.45,
right=0.55,
size=n_neurons * n_ensembles)
intercepts = nengo.dists.CosineSimilarity(n_input + 2).ppf(1 -
triangular)
intercepts = intercepts.reshape((n_ensembles, n_neurons))
# Setting weights for the ensembles -----------------------------------------
# TODO: using presaved weights
weights = np.zeros((self.n_ensembles, n_output, self.n_neurons))
# Setting encoders for the ensembles ----------------------------------------
# if NengoLib is installed, use it to optimize encoder placement
try:
encoders_dist = ScatteredHypersphere(surface=True)
except ImportError:
encoders_dist = nengo.Default
print("NengoLib not installed, encoder placement will " +
"be sub-optimal.")
encoders = encoders_dist.sample(n_neurons * n_ensembles, n_input)
encoders = encoders.reshape(n_ensembles, n_neurons, n_input)
# Defining the Nengo adaptive model ------------------------------------------
self.nengo_model = nengo.Network(seed=seed)
self.nengo_model.config[nengo.Ensemble].neuron_type = nengo.LIF()
with self.nengo_model:
def input_signals_func(t):
return self.input_signal
input_signals = nengo.Node(input_signals_func, size_out=n_input)
def training_signals_func(t):
return -self.training_signal
training_signals = nengo.Node(training_signals_func,
size_out=n_output)
def output_func(t, x):
self.output = np.copy(x)
output = nengo.Node(output_func, size_in=n_output, size_out=0)
self.adapt_ens = []
self.conn_learn = []
for ii in range(self.n_ensembles):
self.adapt_ens.append(
nengo.Ensemble(
n_neurons=self.n_neurons,
dimensions=n_input,
intercepts=intercepts[ii],
radius=np.sqrt(n_input),
encoders=encoders[ii],
**kwargs,
))
# hook up input signal to adaptive population to provide context
nengo.Connection(
input_signals,
self.adapt_ens[ii],
synapse=self.tau_input,
)
self.conn_learn.append(
nengo.Connection(
self.adapt_ens[ii].neurons,
output,
learning_rule_type=nengo.PES(pes_learning_rate),
transform=weights[ii],
synapse=self.tau_output,
))
# hook up the training signal to the learning rule
nengo.Connection(
training_signals,
self.conn_learn[ii].learning_rule,
synapse=self.tau_training,
)
nengo.rc.set("decoder_cache", "enabled", "False")
self.sim = nengo.Simulator(self.nengo_model, dt=0.001)
# weights_probe = nengo.Probe(conn1,"weights",sample_every=probe_sample_rate)
#if(not full_log):
# nengo.Node(log)
#############################
step_time = (presentation_time + pause_time)
Args = {
"Dataset": Dataset,
"Labels": label_train_filtered,
"step_time": step_time,
"input_nbr": input_nbr
}
with nengo.Simulator(model, dt=dt) as sim:
#if(not full_log):
# log.set(sim,Args,False,False)
w.output.set_signal_vmem(
sim.signals[sim.model.sig[input_layer.neurons]["voltage"]])
w.output.set_signal_out(sim.signals[sim.model.sig[layer1.neurons]["out"]])
sim.run(iterations * step_time * label_train_filtered.shape[0])
# weights = sim.data[weights_probe][-1]
weights = weights[-1]
#if(not full_log):
# log.closeLog()
def test_validation(Simulator):
with nengo.Network() as net:
# not a callable
with pytest.raises(ValidationError):
TensorNode([0])
# size out < 1
with pytest.raises(ValidationError):
TensorNode(lambda t: t, size_out=0)
# wrong call signature
with pytest.raises(ValidationError):
TensorNode(lambda a, b, c: a)
# returning None
with pytest.raises(ValidationError):
TensorNode(lambda x: None)
# returning non-tensor
with pytest.raises(ValidationError):
TensorNode(lambda x: [0])
# returning wrong number of dimensions
with pytest.raises(ValidationError):
TensorNode(lambda t: tf.zeros((2, 2, 2)))
# correct output
n = TensorNode(lambda t: tf.zeros((5, 2)))
assert n.size_out == 2
# can't run tensornode in regular Nengo simulator
with nengo.Simulator(net) as sim:
with pytest.raises(SimulationError):
sim.step()
# these tensornodes won't be validated at creation, because size_out
# is specified. instead the validation occurs when the network is built
# None output
with nengo.Network() as net:
TensorNode(lambda t: None, size_out=2)
with pytest.raises(ValidationError):
Simulator(net)
# wrong number of dimensions
with nengo.Network() as net:
TensorNode(lambda t: tf.zeros((1, 2, 2)), size_out=2)
with pytest.raises(ValidationError):
Simulator(net)
# wrong minibatch size
with nengo.Network() as net:
TensorNode(lambda t: tf.zeros((3, 2)), size_out=2)
with pytest.raises(ValidationError):
Simulator(net, minibatch_size=2)
# wrong output d
with nengo.Network() as net:
TensorNode(lambda t: tf.zeros((3, 2)), size_out=3)
with pytest.raises(ValidationError):
Simulator(net, minibatch_size=3)
# wrong dtype
with nengo.Network() as net:
TensorNode(lambda t: tf.zeros((3, 2), dtype=tf.int32), size_out=2)
with pytest.raises(ValidationError):
Simulator(net, minibatch_size=3)
# make sure that correct output _does_ pass
with nengo.Network() as net:
TensorNode(lambda t: tf.zeros((3, 2), dtype=t.dtype), size_out=2)
with Simulator(net, minibatch_size=3):
pass
N = 10
def comparator_func(t, x):
R1 = np.correlate(x[:N], x[N:])
print(R1)
return R1
with nengo.Network() as net:
ens1 = nengo.Ensemble(10,
dimensions=1,
seed=0,
intercepts=nengo.dists.Choice([-0.1]),
max_rates=nengo.dists.Choice([100]))
ens2 = nengo.Ensemble(10,
dimensions=1,
seed=0,
intercepts=nengo.dists.Choice([-0.1]),
max_rates=nengo.dists.Choice([100]))
node = nengo.Node(size_in=20, output=comparator_func)
# Neuron to neuron
weights = np.eye(ens1.n_neurons, ens1.n_neurons)
nengo.Connection(ens1.neurons, node[:10], transform=weights)
nengo.Connection(ens2.neurons, node[10:], transform=weights)
with nengo.Simulator(net) as sim:
sim.run(0.1)
def evaluate_mnist_multiple_local(args):
#############################
# load the data
#############################
input_nbr = args.input_nbr
(image_train,
label_train), (image_test,
label_test) = (keras.datasets.mnist.load_data())
probe_sample_rate = (
input_nbr / 10
) / 1000 #Probe sample rate. Proportional to input_nbr to scale down sampling rate of simulations
# # probe_sample_rate = 1000
image_train_filtered = []
label_train_filtered = []
x = args.digit
for i in range(0, input_nbr):
image_train_filtered.append(image_train[i])
label_train_filtered.append(label_train[i])
image_train_filtered = np.array(image_train_filtered)
label_train_filtered = np.array(label_train_filtered)
image_train_filtered = (image_train_filtered / 255 - 0.1308) / 0.3088
image_train_filtered = 255 * (
image_train_filtered - image_train_filtered.min()) / (
image_train_filtered.max() - image_train_filtered.min())
# np.save(
# 'mnist.npz',
# image_train_filtered=image_train_filtered,
# label_train_filtered=label_train_filtered,
# image_test_filtered=image_test_filtered,
# label_test_filtered=label_test_filtered,
# )
# data = np.load('qmnist.npz', allow_pickle=True)
# image_train_filtered = data['image_train_filtered']
# label_train_filtered = data['label_train_filtered']
# image_test_filtered = data['image_test_filtered']
# label_test_filtered = data['label_test_filtered']
# image_train_filtered = np.tile(image_train_filtered,(args.iterations,1,1))
# label_train_filtered = np.tile(label_train_filtered,(args.iterations))
#Simulation Parameters
#Presentation time
presentation_time = args.presentation_time #0.20
#Pause time
pause_time = args.pause_time
#Iterations
iterations = args.iterations
#Input layer parameters
n_in = args.n_in
# g_max = 1/784 #Maximum output contribution
n_neurons = args.n_neurons # Layer 1 neurons
# inhib_factor = args.inhib_factor #Multiplication factor for lateral inhibition
input_neurons_args = {
"n_neurons":
n_in,
"dimensions":
1,
"label":
"Input layer",
"encoders":
nengo.dists.Choice([[1]]),
# "max_rates":nengo.dists.Uniform(22,22),
# "intercepts":nengo.dists.Uniform(0,0),
"gain":
nengo.dists.Choice([args.gain_in]),
"bias":
nengo.dists.Choice([args.bias_in]),
"neuron_type":
MyLIF_in(tau_rc=args.tau_in, min_voltage=-1, amplitude=args.amp_neuron)
# "neuron_type":nengo.neurons.SpikingRectifiedLinear()#SpikingRelu neuron.
}
#Layer 1 parameters
layer_1_neurons_args = {
"n_neurons":
n_neurons,
"dimensions":
1,
"label":
"Layer 1",
"encoders":
nengo.dists.Choice([[1]]),
"gain":
nengo.dists.Choice([args.gain_out]),
"bias":
nengo.dists.Choice([args.bias_out]),
# "intercepts":nengo.dists.Choice([0]),
# "max_rates":nengo.dists.Choice([args.rate_out,args.rate_out]),
# "noise":nengo.processes.WhiteNoise(dist=nengo.dists.Gaussian(0, 0.5), seed=1),
# "neuron_type":nengo.neurons.LIF(tau_rc=args.tau_out, min_voltage=0)
# "neuron_type":MyLIF_out(tau_rc=args.tau_out, min_voltage=-1)
"neuron_type":
STDPLIF(tau_rc=args.tau_out,
min_voltage=-1,
spiking_threshold=args.thr_out,
inhibition_time=args.inhibition_time)
}
# "noise":nengo.processes.WhiteNoise(dist=nengo.dists.Gaussian(0, 20), seed=1),
#Lateral Inhibition parameters
# lateral_inhib_args = {
# "transform": inhib_factor* (np.full((n_neurons, n_neurons), 1) - np.eye(n_neurons)),
# "synapse":args.inhib_synapse,
# "label":"Lateral Inhibition"
# }
#Learning rule parameters
learning_args = {
"lr": args.lr,
"winit_min": 0,
"winit_max": 1,
"vprog": args.vprog,
"vthp": args.vthp,
"vthn": args.vthn,
# "tpw":50,
# "prev_flag":True,
"sample_distance": int((presentation_time + pause_time) * 200 *
10), #Store weight after 10 images
}
# argument_string = "presentation_time: "+ str(presentation_time)+ "\n pause_time: "+ str(pause_time)+ "\n input_neurons_args: " + str(input_neurons_args)+ " \n layer_1_neuron_args: " + str(layer_1_neurons_args)+"\n Lateral Inhibition parameters: " + str(lateral_inhib_args) + "\n learning parameters: " + str(learning_args)+ "\n g_max: "+ str(g_max)
images = image_train_filtered
labels = label_train_filtered
model = nengo.Network("My network", seed=args.seed)
#############################
# Model construction
#############################
with model:
# picture = nengo.Node(PresentInputWithPause(images, presentation_time, pause_time,0))
picture = nengo.Node(
nengo.processes.PresentInput(images,
presentation_time=presentation_time))
true_label = nengo.Node(
nengo.processes.PresentInput(labels,
presentation_time=presentation_time))
# true_label = nengo.Node(PresentInputWithPause(labels, presentation_time, pause_time,-1))
# input layer
input_layer = nengo.Ensemble(**input_neurons_args)
input_conn = nengo.Connection(picture,
input_layer.neurons,
synapse=None)
#first layer
layer1 = nengo.Ensemble(**layer_1_neurons_args)
#Weights between input layer and layer 1
w = nengo.Node(CustomRule_post_v2(**learning_args),
size_in=n_in,
size_out=n_neurons)
nengo.Connection(input_layer.neurons, w, synapse=None)
nengo.Connection(w, layer1.neurons, synapse=None)
# nengo.Connection(w, layer1.neurons,transform=g_max, synapse=None)
init_weights = np.random.uniform(0, 1, (n_neurons, n_in))
# conn1 = nengo.Connection(input_layer.neurons,layer1.neurons,learning_rule_type=VLR(learning_rate=args.lr,vprog=args.vprog, vthp=args.vthp,vthn=args.vthn),transform=init_weights)
#Lateral inhibition
# inhib = nengo.Connection(layer1.neurons,layer1.neurons,**lateral_inhib_args)
#Probes
p_true_label = nengo.Probe(true_label, sample_every=probe_sample_rate)
p_input_layer = nengo.Probe(input_layer.neurons,
sample_every=probe_sample_rate)
p_layer_1 = nengo.Probe(layer1.neurons, sample_every=probe_sample_rate)
# weights_probe = nengo.Probe(conn1,"weights",sample_every=probe_sample_rate)
weights = w.output.history
# with nengo_ocl.Simulator(model) as sim :
with nengo.Simulator(model, dt=args.dt, optimize=True) as sim:
w.output.set_signal_vmem(
sim.signals[sim.model.sig[input_layer.neurons]["voltage"]])
w.output.set_signal_out(
sim.signals[sim.model.sig[layer1.neurons]["out"]])
sim.run((presentation_time + pause_time) * labels.shape[0])
#save the model
# now = time.strftime("%Y%m%d-%H%M%S")
# folder = os.getcwd()+"/MNIST_VDSP"+now
# os.mkdir(folder)
# print(weights)
# weights = sim.data[weights_probe]
last_weight = weights[-1]
# pickle.dump(weights, open( folder+"/trained_weights", "wb" ))
# pickle.dump(argument_string, open( folder+"/arguments", "wb" ))
t_data = sim.trange(sample_every=probe_sample_rate)
labels = sim.data[p_true_label][:, 0]
output_spikes = sim.data[p_layer_1]
neuron_class = np.zeros((n_neurons, 1))
n_classes = 10
for j in range(n_neurons):
spike_times_neuron_j = t_data[np.where(output_spikes[:, j] > 0)]
max_spike_times = 0
for i in range(n_classes):
class_presentation_times_i = t_data[np.where(labels == i)]
#Normalized number of spikes wrt class presentation time
num_spikes = len(
np.intersect1d(spike_times_neuron_j,
class_presentation_times_i)) / (
len(class_presentation_times_i) + 1)
if (num_spikes > max_spike_times):
neuron_class[j] = i
max_spike_times = num_spikes
# print("Neuron class: \n", neuron_class)
sim.close()
'''
Testing
'''
# img_rows, img_cols = 28, 28
input_nbr = 6000
# input_nbr = int(args.input_nbr/6)
# Dataset = "Mnist"
# # (image_train, label_train), (image_test, label_test) = load_mnist()
# (image_train, label_train), (image_test, label_test) = (tf.keras.datasets.mnist.load_data())
# #select the 0s and 1s as the two classes from MNIST data
image_test_filtered = []
label_test_filtered = []
for i in range(0, input_nbr):
# # if (label_train[i] == 1 or label_train[i] == 0):
image_test_filtered.append(image_test[i])
label_test_filtered.append(label_test[i])
print("actual input", len(label_test_filtered))
print(np.bincount(label_test_filtered))
image_test_filtered = np.array(image_test_filtered)
label_test_filtered = np.array(label_test_filtered)
image_test_filtered = (image_test_filtered / 255 - 0.1308) / 0.3088
image_test_filtered = 255 * (
image_test_filtered - image_test_filtered.min()) / (
image_test_filtered.max() - image_test_filtered.min())
#############################
model = nengo.Network(label="My network", )
# Learning params
with model:
# input layer
# picture = nengo.Node(PresentInputWithPause(images, presentation_time, pause_time,0))
picture = nengo.Node(
nengo.processes.PresentInput(image_test_filtered,
presentation_time=presentation_time))
true_label = nengo.Node(
nengo.processes.PresentInput(label_test_filtered,
presentation_time=presentation_time))
# true_label = nengo.Node(PresentInputWithPause(labels, presentation_time, pause_time,-1))
input_layer = nengo.Ensemble(**input_neurons_args)
input_conn = nengo.Connection(picture,
input_layer.neurons,
synapse=None)
#first layer
layer1 = nengo.Ensemble(**layer_1_neurons_args)
# w = nengo.Node(CustomRule_post_v2(**learning_args), size_in=784, size_out=n_neurons)
nengo.Connection(input_layer.neurons,
layer1.neurons,
transform=last_weight)
p_true_label = nengo.Probe(true_label)
p_layer_1 = nengo.Probe(layer1.neurons)
p_input_layer = nengo.Probe(input_layer.neurons)
#if(not full_log):
# nengo.Node(log)
#############################
step_time = (presentation_time + pause_time)
with nengo.Simulator(model, dt=args.dt) as sim:
sim.run(step_time * label_test_filtered.shape[0])
labels = sim.data[p_true_label][:, 0]
output_spikes = sim.data[p_layer_1]
n_classes = 10
# rate_data = nengo.synapses.Lowpass(0.1).filtfilt(sim.data[p_layer_1])
predicted_labels = []
true_labels = []
correct_classified = 0
wrong_classified = 0
class_spikes = np.ones((10, 1))
for num in range(input_nbr):
#np.sum(sim.data[my_spike_probe] > 0, axis=0)
output_spikes_num = output_spikes[
num * int(presentation_time / args.dt):(num + 1) *
int(presentation_time / args.dt), :] # 0.350/0.005
num_spikes = np.sum(output_spikes_num > 0, axis=0)
for i in range(n_classes):
sum_temp = 0
count_temp = 0
for j in range(n_neurons):
if ((neuron_class[j]) == i):
sum_temp += num_spikes[j]
count_temp += 1
if (count_temp == 0):
class_spikes[i] = 0
else:
class_spikes[i] = sum_temp
# class_spikes[i] = sum_temp/count_temp
# print(class_spikes)
k = np.argmax(num_spikes)
# predicted_labels.append(neuron_class[k])
class_pred = np.argmax(class_spikes)
predicted_labels.append(class_pred)
true_class = labels[(num * int(presentation_time / args.dt))]
# print(true_class)
# print(class_pred)
# if(neuron_class[k] == true_class):
# correct_classified+=1
# else:
# wrong_classified+=1
if (class_pred == true_class):
correct_classified += 1
else:
wrong_classified += 1
accuracy = correct_classified / (correct_classified +
wrong_classified) * 100
print("Accuracy: ", accuracy)
sim.close()
# nni.report_final_result(accuracy)
del weights, sim.data, labels, output_spikes, class_pred, t_data
return accuracy, last_weight
def test_conv_split(Simulator, rng, plt, allclose):
channels_last = False
# load data
with open(os.path.join(test_dir, 'mnist10.pkl'), 'rb') as f:
test10 = pickle.load(f)
input_shape = ImageShape(28, 28, 1, channels_last=channels_last)
test_x = test10[0][0].reshape(input_shape.shape(channels_last=True))
if not input_shape.channels_last:
test_x = np.transpose(test_x, (2, 0, 1))
n_filters = 8
kernel_size = (7, 7)
kernel = Gabor(freq=Uniform(0.5, 1)).generate(n_filters,
kernel_size,
rng=rng)
kernel = kernel[None, :, :, :] # single channel
kernel = np.transpose(kernel, (0, 2, 3, 1)) # filters last
strides = (2, 2)
seed = 3 # fix seed to do the same computation for both channel positions
rng = np.random.RandomState(seed + 1)
with nengo.Network(seed=seed) as net:
nengo_loihi.add_params(net)
a = nengo.Node(test_x.ravel())
# --- make population to turn image into spikes
nc = 1
in_kernel = np.array([1.]).reshape((1, 1, 1, nc))
transform = nengo_loihi.Conv2D.from_kernel(in_kernel, input_shape)
b = nengo.Ensemble(transform.output_shape.size,
1,
neuron_type=nengo.SpikingRectifiedLinear(),
max_rates=nengo.dists.Choice([50]),
intercepts=nengo.dists.Choice([0]))
net.config[b].on_chip = False
nengo.Connection(a, b.neurons, transform=transform)
in_shape = transform.output_shape
transform = nengo_loihi.Conv2D.from_kernel(kernel,
in_shape,
strides=strides)
out_shape = transform.output_shape
split_slices = out_shape.split_channels(max_size=1024, max_channels=4)
# --- make convolution population, split across ensembles
cc = []
cp = []
out_shapes = []
xslice = ImageSlice(in_shape)
for yslice in split_slices:
transform_xy = split_transform(transform, xslice, yslice)
out_shapes.append(transform_xy.output_shape)
c = nengo.Ensemble(transform_xy.output_shape.size,
1,
neuron_type=nengo.LIF(),
max_rates=nengo.dists.Choice([15]),
intercepts=nengo.dists.Choice([0]))
nengo.Connection(b.neurons, c.neurons, transform=transform_xy)
cc.append(c)
cp.append(nengo.Probe(c.neurons))
with nengo.Simulator(net, optimize=False) as sim_nengo:
sim_nengo.run(1.0)
with Simulator(net, seed=seed) as sim_loihi:
if "loihi" in sim_loihi.sims:
sim_loihi.sims["loihi"].snip_max_spikes_per_step = 100
sim_loihi.run(1.0)
nengo_out = []
loihi_out = []
for p, out_shape_i in zip(cp, out_shapes):
nengo_out.append(
(sim_nengo.data[p] > 0).sum(axis=0).reshape(out_shape_i.shape()))
loihi_out.append(
(sim_loihi.data[p] > 0).sum(axis=0).reshape(out_shape_i.shape()))
if channels_last:
nengo_out = np.concatenate(nengo_out, axis=2)
loihi_out = np.concatenate(loihi_out, axis=2)
# put channels first to display them separately
nengo_out = np.transpose(nengo_out, (2, 0, 1))
loihi_out = np.transpose(loihi_out, (2, 0, 1))
else:
nengo_out = np.concatenate(nengo_out, axis=0)
loihi_out = np.concatenate(loihi_out, axis=0)
out_max = np.maximum(nengo_out.max(), loihi_out.max())
# --- plot results
rows = 2
cols = 3
ax = plt.subplot(rows, cols, 1)
imshow(test_x[0, :, :], vmin=0, vmax=1, ax=ax)
ax = plt.subplot(rows, cols, 2)
tile(np.transpose(kernel[0], (2, 0, 1)), cols=8, ax=ax)
ax = plt.subplot(rows, cols, 3)
plt.hist(nengo_out.ravel(), bins=31)
plt.hist(loihi_out.ravel(), bins=31)
ax = plt.subplot(rows, cols, 4)
tile(nengo_out, vmin=0, vmax=out_max, cols=8, ax=ax)
ax = plt.subplot(rows, cols, 6)
tile(loihi_out, vmin=0, vmax=out_max, cols=8, ax=ax)
assert allclose(loihi_out, nengo_out, atol=0.05 * out_max, rtol=0.15)
def plot_tuning_curves(filename, plot_decoding=False, show=False):
"""
Plot tuning curves for an association population and for a standard
subpopulation (of the neural extraction network).
"""
import matplotlib as mpl
mpl.rcParams['font.size'] = '10'
if show:
mpl.use('Qt4Agg')
else:
mpl.use('Agg')
import matplotlib.pyplot as plt
plt.figure(figsize=(5, 3))
neurons_per_item = 20
neurons_per_dim = 50
intercepts_low = 0.29
intercepts_range = 0.00108
intercepts = Uniform(intercepts_low, intercepts_low + intercepts_range)
tau_rc = 0.034
tau_ref = 0.0026
radius = 1.0
assoc_encoders = np.ones((neurons_per_item, 1))
standard_encoders = np.ones((neurons_per_dim, 1))
threshold = 0.3
threshold_func = lambda x: 1 if x > threshold else 0
max_rates = Uniform(200, 350)
model = nengo.Network("Associative Memory")
with model:
neuron_type = nengo.LIF(tau_rc=tau_rc, tau_ref=tau_ref)
assoc = nengo.Ensemble(n_neurons=neurons_per_item,
dimensions=1,
intercepts=intercepts,
encoders=assoc_encoders,
label="assoc",
radius=radius,
max_rates=max_rates,
neuron_type=neuron_type)
n_eval_points = 750
eval_points = np.random.normal(0, 0.06, (n_eval_points, 1))
eval_points.T[0].sort()
radius = 5.0 / np.sqrt(512)
standard = nengo.Ensemble(n_neurons=neurons_per_dim,
dimensions=1,
eval_points=eval_points,
radius=radius,
encoders=standard_encoders)
if plot_decoding:
dummy = nengo.Ensemble(1, 1)
conn = nengo.Connection(assoc, dummy, function=threshold_func)
dummy2 = nengo.Ensemble(1, 1)
conn2 = nengo.Connection(standard, dummy2)
sim = nengo.Simulator(model)
if plot_decoding:
gs = gridspec.GridSpec(3, 2)
else:
gs = gridspec.GridSpec(2, 2)
plt.subplot(gs[0:2, 0])
assoc_eval_points, assoc_activities = tuning_curves(assoc, sim)
for neuron in assoc_activities.T:
plt.plot(assoc_eval_points.T[0], neuron)
plt.title("Association")
plt.ylabel("Firing Rate (spikes/s)")
plt.xlabel(r"$e_ix$")
plt.ylim((0, 400))
plt.yticks([0, 100, 200, 300, 400])
ax = plt.subplot(gs[0:2, 1])
# We want different eval points for display purposes than for
# optimization purposes
eval_points = Uniform(-radius, radius).sample(n_eval_points)
eval_points.sort()
eval_points = eval_points.reshape((n_eval_points, 1))
# have to divide by radius on our own since tuning_curves skips that step
_, activities = tuning_curves(standard, sim, eval_points / radius)
for neuron in activities.T:
plt.plot(eval_points, neuron)
plt.title("Standard")
plt.xlabel(r"$e_ix$")
plt.xlim((-radius, radius))
plt.ylim((0, 400))
plt.setp(ax, yticks=[])
if plot_decoding:
plt.subplot(gs[2, 0])
decoders = sim.data[conn].decoders
plt.plot(assoc_eval_points.T[0],
0.001 * np.dot(assoc_activities, decoders.T))
plt.axhline(y=1.0, ls='--')
plt.subplot(gs[2, 1])
x, activities2 = tuning_curves(standard, sim,
assoc_eval_points / radius)
decoders = sim.data[conn2].decoders
plt.plot(assoc_eval_points.T[0],
0.001 * np.dot(activities2, decoders.T))
plt.plot([-1.0, 1.0], [-1.0, 1.0], c='k', ls='--')
plt.axvline(x=radius, c='k', ls='--')
plt.axvline(x=-radius, c='k', ls='--')
plt.tight_layout()
plt.subplots_adjust(right=0.89, left=0.11)
if filename:
plt.savefig(filename)
if show:
plt.show()
np.save("y_spikes", sim.data[y_spikes])'''
#fig=plt.figure()
#plt.subplot(2, 1, 1)
#plt.plot(sim.trange(), sim.data[x_spikes])
#plt.subplot(2, 1, 2)
#plot_spikes(sim.trange(), sim.data[y_spikes])
#plt.xlabel("Time [s]")
#plt.ylabel("Neuron number")
#plt.savefig('fig.png')
if __name__ == '__main__':
print("starting simulator...")
before = time.time()
sim = nengo.Simulator(model)
after = time.time()
print("time to build:")
print(after - before)
print("running simulator...")
before = time.time()
while True:
sim.step()
time.sleep(0.0001)
def __init__( # noqa: C901
self,
network,
dt=0.001,
seed=None,
model=None,
precompute=None,
target=None,
progress_bar=None,
remove_passthrough=True,
hardware_options=None,
):
# initialize values used in __del__ and close() first
self.closed = True
self.network = network
self.sims = OrderedDict()
self.timers = Timers()
self.timers.start("build")
self.seed = seed
self._n_steps = 0
self._time = 0
hardware_options = {} if hardware_options is None else hardware_options
if progress_bar:
warnings.warn("nengo-loihi does not support progress bars")
if model is None:
self.model = Model(dt=float(dt), label="%s, dt=%f" % (network, dt))
else:
assert isinstance(
model, Model), "model is not type 'nengo_loihi.builder.Model'"
self.model = model
assert self.model.dt == dt
if network is None:
raise ValidationError("network parameter must not be None",
attr="network")
if target is None:
target = "loihi" if HAS_NXSDK else "sim"
self.target = target
logger.info("Simulator target is %r", target)
# Build the network into the model
self.model.build(
network,
precompute=precompute,
remove_passthrough=remove_passthrough,
discretize=target != "simreal",
)
# Create host_pre and host simulators if necessary
self.precompute = self.model.split.precompute
logger.info("Simulator precompute is %r", self.precompute)
assert precompute is None or precompute == self.precompute
if self.model.split.precomputable() and not self.precompute:
warnings.warn(
"Model is precomputable. Setting precompute=False may slow execution."
)
if len(self.model.host_pre.params) > 0:
assert self.precompute
self.sims["host_pre"] = nengo.Simulator(
network=None,
dt=self.dt,
model=self.model.host_pre,
progress_bar=False,
optimize=False,
)
if len(self.model.host.params) > 0:
self.sims["host"] = nengo.Simulator(
network=None,
dt=self.dt,
model=self.model.host,
progress_bar=False,
optimize=False,
)
self._probe_outputs = self.model.params
self.data = SimulationData(self._probe_outputs)
for sim in self.sims.values():
self.data.add_fallback(sim.data)
if seed is None:
if network is not None and network.seed is not None:
seed = network.seed + 1
else:
seed = np.random.randint(npext.maxint)
if target in ("simreal", "sim"):
self.sims["emulator"] = EmulatorInterface(self.model, seed=seed)
elif target == "loihi":
assert HAS_NXSDK, "Must have NxSDK installed to use Loihi hardware"
use_snips = not self.precompute and self.sims.get("host",
None) is not None
self.sims["loihi"] = HardwareInterface(self.model,
use_snips=use_snips,
seed=seed,
**hardware_options)
else:
raise ValidationError("Must be 'simreal', 'sim', or 'loihi'",
attr="target")
assert "emulator" in self.sims or "loihi" in self.sims
self._runner = StepRunner(self.model, self.sims, self.precompute,
self.timers)
self.closed = False
self.timers.stop("build")