# +-------------------------------------------------------------------+
# Record neurons' potentials
pre_pop.record(['v', 'spikes'])
post_pop.record(['v', 'spikes'])
# Run simulation
sim.run(simtime)
print("Weights:{}".format(plastic_projection.get('weight', 'list')))
pre_spikes = pre_pop.get_data('spikes')
post_spikes = post_pop.get_data('spikes')
Figure(
# raster plot of the presynaptic neuron spike times
Panel(pre_spikes.segments[0].spiketrains,
yticks=True,
markersize=0.2,
xlim=(0, simtime)),
Panel(post_spikes.segments[0].spiketrains,
yticks=True,
markersize=0.2,
xlim=(0, simtime)),
title="stdp example curr",
annotations="Simulated with {}".format(sim.name()))
plt.show()
# End simulation on SpiNNaker
sim.end()
def lancement_sim(cellSourceSpikes,
path,
weight_input=0,
weight_inter=0,
max_time=800000,
TIME_STEP=TIME_STEP,
input_n=input_n,
nb_neuron_int=nb_neuron_int,
nb_neuron_out=nb_neuron_out,
delay=delay,
p_conn_in_int=p_conn_in_int,
p_conn_int_out=p_conn_int_out,
v_tresh=v_tresh):
simulator = 'spinnaker'
# le max_delay doit être inférieur à 14*time_step
sim.setup(timestep=TIME_STEP, min_delay=delay, max_delay=delay * 2)
randoms = np.random.rand(100, 1)
#defining network topology
lif_curr_exp_params = {
'cm': 1.0, # The capacitance of the LIF neuron in nano-Farads
'tau_m': 20.0, # The time-constant of the RC circuit, in milliseconds
'tau_refrac': 5.0, # The refractory period, in milliseconds
'v_reset':
-65.0, # The voltage to set the neuron at immediately after a spike
'v_rest': -65.0, # The ambient rest voltage of the neuron
'v_thresh':
-(v_tresh), # The threshold voltage at which the neuron will spike
'tau_syn_E': 5.0, # The excitatory input current decay time-constant
'tau_syn_I': 5.0, # The inhibitory input current decay time-constant
'i_offset': 0.0, # A base input current to add each timestep
}
# Population d'entrée avec comme source le SpikeSourceArray en paramètre
Input = sim.Population(input_n,
sim.SpikeSourceArray(spike_times=cellSourceSpikes),
label="Input")
Input.record("spikes")
# Définition des types de neurones et des couches intermédiaire, de sortie, ainsi que celle contenant le neurone de l'attention
LIF_Intermediate = sim.IF_curr_exp(**lif_curr_exp_params)
Intermediate = sim.Population(nb_neuron_int,
LIF_Intermediate,
label="Intermediate")
Intermediate.record(("spikes", "v"))
LIF_Output = sim.IF_curr_exp(**lif_curr_exp_params)
Output = sim.Population(nb_neuron_out, LIF_Output, label="Output")
Output.record(("spikes", "v"))
LIF_delayer = sim.IF_curr_exp(**lif_curr_exp_params)
Delay_n = sim.Population(1, LIF_delayer, label="Delay")
Delay_n.record(("spikes", "v"))
# set the stdp mechanisim parameters, we are going to use stdp in both connections between (input-intermediate) adn (intermediate-output)
python_rng = NumpyRNG(seed=98497627)
delay = delay # (ms) synaptic time delay
#A_minus
# définition des connexions entre couches de neurones entrée <=> intermédiaire, intermédiaire <=> sortie
# vérificatio pour savoir si on est dans le cas de la première simulation par défault ou si on doit injecter les poids
if ((weight_input != 0) or (weight_inter != 0)):
# cas ou l'on inject les poids
Conn_input_inter = sim.Projection(
Input,
Intermediate,
# Le fromListConnector pour injecter les poids
connector=sim.FromListConnector(weight_input),
receptor_type="excitatory",
label="Connection input to intermediate",
# des synapses static pour "suprimer" l'apprentissage
synapse_type=sim.StaticSynapse())
Conn_inter_output = sim.Projection(
Intermediate,
Output, # pre and post population
connector=sim.FromListConnector(weight_inter),
receptor_type="excitatory",
label="Connection intermediate to output",
synapse_type=sim.StaticSynapse())
else:
# cas par défault
Conn_input_inter = sim.Projection(
Input,
Intermediate,
connector=sim.FixedProbabilityConnector(
p_conn_in_int, allow_self_connections=False),
synapse_type=sim.StaticSynapse(
weight=RandomDistribution('normal', (3, 2.9), rng=python_rng)),
receptor_type="excitatory",
label="Connection input to intermediate")
Conn_inter_output = sim.Projection(
Intermediate,
Output, # pre and post population
connector=sim.FixedProbabilityConnector(
p_conn_int_out, allow_self_connections=False),
synapse_type=sim.StaticSynapse(
weight=RandomDistribution('normal', (3, 2.9), rng=python_rng)),
receptor_type="excitatory",
label="Connection intermediate to output")
# définition des connexions inhibitrices des couches intermédiaire et de sortie
FixedInhibitory_WTA = sim.StaticSynapse(weight=6)
WTA_INT = sim.Projection(
Intermediate,
Intermediate,
connector=sim.AllToAllConnector(allow_self_connections=False),
synapse_type=FixedInhibitory_WTA,
receptor_type="inhibitory",
label="Connection WTA")
WTA_OUT = sim.Projection(
Output,
Output,
connector=sim.AllToAllConnector(allow_self_connections=False),
synapse_type=FixedInhibitory_WTA,
receptor_type="inhibitory",
label="Connection WTA")
# Connexion avec le neurone de l'attention
FixedInhibitory_delayer = sim.StaticSynapse(weight=2)
Delay_out = sim.Projection(
Delay_n,
Output,
connector=sim.AllToAllConnector(allow_self_connections=False),
synapse_type=FixedInhibitory_delayer,
receptor_type="inhibitory",
label="Connection WTA")
Delay_inter = sim.Projection(
Intermediate,
Delay_n,
connector=sim.AllToAllConnector(allow_self_connections=False),
synapse_type=FixedInhibitory_delayer,
receptor_type="inhibitory",
label="Connection WTA")
# On précise le nombre de neurone par coeurs au cas ou
sim.set_number_of_neurons_per_core(sim.IF_curr_exp, 255)
# on arrondie le temps de simulation, sinon avec les callbacks, on a une boucle infinie pour des temps d'arrêts plus précis que la fréquence des callbacks
simtime = ceil(max_time)
try:
#lancement de la simulation
sim.run(simtime)
#récupération des infos sur les spike des trois couches
neo = Output.get_data(variables=["spikes", "v"])
spikes = neo.segments[0].spiketrains
#print(spikes)
v = neo.segments[0].filter(name='v')[0]
neo_in = Input.get_data(variables=["spikes"])
spikes_in = neo_in.segments[0].spiketrains
#print(spikes_in)
neo_intermediate = Intermediate.get_data(variables=["spikes", "v"])
spikes_intermediate = neo_intermediate.segments[0].spiketrains
#print(spikes_intermediate)
v_intermediate = neo_intermediate.segments[0].filter(name='v')[0]
#print(v_intermediate)
sim.reset()
sim.end()
except:
# Si la simulation fail, on set ces deux variables à zéros pour gérer l'erreur dans le script principal
v = 0
spikes = 0
# Création et sauvegarde des graphs des graphs si la simluation s'est bien passée, + envoie des sorties de la fonction
if (isinstance(spikes, list) and isinstance(v, AnalogSignal)):
plot.Figure(
# plot voltage for first ([0]) neuron
plot.Panel(v,
ylabel="Membrane potential (mV)",
data_labels=[Output.label],
yticks=True,
xlim=(0, simtime)),
# plot spikes (or in this case spike)
plot.Panel(spikes, yticks=True, markersize=5, xlim=(0, simtime)),
title="Spiking activity of the output layer during test",
annotations="Simulated with {}".format(sim.name())).save(
"./Generated_data/tests/" + path +
"/output_layer_membrane_voltage_and_spikes.png")
plot.Figure(
# plot voltage for first ([0]) neuron
plot.Panel(v_intermediate,
ylabel="Membrane potential (mV)",
data_labels=[Output.label],
yticks=True,
xlim=(0, simtime)),
# plot spikes (or in this case spike)
plot.Panel(spikes_intermediate,
yticks=True,
markersize=5,
xlim=(0, simtime)),
title="Spiking activity of the intermediate layer during test",
annotations="Simulated with {}".format(sim.name())).save(
"./Generated_data/tests/" + path +
"/intermediate_layer_membrane_voltage_and_spikes.png")
return v, spikes
else:
print(
"simulation failed with parameters : (l'affichage des paramètres ayant causés le disfonctionnement de la simulation sera traitée à une date ultérieur, merci!)"
)
return 0, 0
path = './results/{}/'.format(int(time.time()))
for i in range(len(pops)):
plot.Figure(
# plot voltage for first ([0]) neuron
plot.Panel(v[i],
ylabel="Membrane potential (mV)",
data_labels=[pops[i].label],
yticks=True,
xlim=(0, SIMTIME)),
# plot spikes (or in this case spike)
plot.Panel(spikes[i], yticks=True, markersize=3, xlim=(0, SIMTIME)),
title="Simple Example",
annotations="Simulated with {}".format(
sim.name())).save(path + 'figure_{}.png'.format(i))
plt.show()
np.savez(file='inputspikes', arr_0=spikes[0])
np.savez(file='pot1', arr_0=v[1])
np.savez(file='pot2', arr_0=v[2])
output_spikes = spikes[2]
output_spike_counts = [
output_spikes[i].size for i in range(len(output_spikes))
]
prediction = np.argmax(output_spike_counts)
print("PREDICTION: {}".format(prediction))
def lancement_sim(cellSourceSpikes,
max_time=800000,
path="default",
TIME_STEP=TIME_STEP,
input_n=input_n,
nb_neuron_int=nb_neuron_int,
nb_neuron_out=nb_neuron_out,
delay=delay,
p_conn_in_int=p_conn_in_int,
p_conn_int_out=p_conn_int_out,
a_minus=0.6,
a_plus=0.6,
tau_minus=12.0,
tau_plus=10.0,
v_tresh=10.0):
simulator = 'spinnaker'
sim.setup(timestep=TIME_STEP, min_delay=delay, max_delay=delay * 2)
randoms = np.random.rand(100, 1)
lif_curr_exp_params = {
'cm': 1.0, # The capacitance of the LIF neuron in nano-Farads
'tau_m': 20.0, # The time-constant of the RC circuit, in milliseconds
'tau_refrac': 5.0, # The refractory period, in milliseconds
'v_reset':
-65.0, # The voltage to set the neuron at immediately after a spike
'v_rest': -65.0, # The ambient rest voltage of the neuron
'v_thresh':
-(v_tresh), # The threshold voltage at which the neuron will spike
'tau_syn_E': 5.0, # The excitatory input current decay time-constant
'tau_syn_I': 5.0, # The inhibitory input current decay time-constant
'i_offset': 0.0, # A base input current to add each timestep
}
Input = sim.Population(input_n,
sim.SpikeSourceArray(spike_times=cellSourceSpikes),
label="Input")
Input.record("spikes")
LIF_Intermediate = sim.IF_curr_exp(**lif_curr_exp_params)
Intermediate = sim.Population(nb_neuron_int,
LIF_Intermediate,
label="Intermediate")
Intermediate.record(("spikes", "v"))
LIF_Output = sim.IF_curr_exp(**lif_curr_exp_params)
Output = sim.Population(nb_neuron_out, LIF_Output, label="Output")
Output.record(("spikes", "v"))
LIF_delayer = sim.IF_curr_exp(**lif_curr_exp_params)
Delay_n = sim.Population(1, LIF_delayer, label="Delay")
Delay_n.record(("spikes", "v"))
python_rng = NumpyRNG(seed=98497627)
delay = delay # (ms) synaptic time delay
# Définition du fonctionnement de la stdp
stdp_proj = sim.STDPMechanism(
timing_dependence=sim.SpikePairRule(tau_plus=tau_plus,
tau_minus=tau_minus,
A_plus=a_plus,
A_minus=a_minus),
weight_dependence=sim.AdditiveWeightDependence(w_min=0.1, w_max=6),
weight=RandomDistribution('normal', (3, 2.9), rng=python_rng),
delay=delay)
Conn_input_inter = sim.Projection(
Input,
Intermediate,
connector=sim.FixedProbabilityConnector(p_conn_in_int,
allow_self_connections=False),
# synapse type set avec la définition de la stdp pour l'aprentsissage
synapse_type=stdp_proj,
receptor_type="excitatory",
label="Connection input to intermediate")
# second projection with stdp
Conn_inter_output = sim.Projection(
Intermediate,
Output, # pre and post population
connector=sim.FixedProbabilityConnector(p_conn_int_out,
allow_self_connections=False),
synapse_type=stdp_proj,
receptor_type="excitatory",
label="Connection intermediate to output")
FixedInhibitory_WTA = sim.StaticSynapse(weight=6)
WTA_INT = sim.Projection(
Intermediate,
Intermediate,
connector=sim.AllToAllConnector(allow_self_connections=False),
synapse_type=FixedInhibitory_WTA,
receptor_type="inhibitory",
label="Connection WTA")
WTA_OUT = sim.Projection(
Output,
Output,
connector=sim.AllToAllConnector(allow_self_connections=False),
synapse_type=FixedInhibitory_WTA,
receptor_type="inhibitory",
label="Connection WTA")
FixedInhibitory_delayer = sim.StaticSynapse(weight=2)
Delay_out = sim.Projection(
Delay_n,
Output,
connector=sim.AllToAllConnector(allow_self_connections=False),
synapse_type=FixedInhibitory_delayer,
receptor_type="inhibitory",
label="Connection WTA")
Delay_inter = sim.Projection(
Intermediate,
Delay_n,
connector=sim.AllToAllConnector(allow_self_connections=False),
synapse_type=FixedInhibitory_delayer,
receptor_type="inhibitory",
label="Connection WTA")
sim.set_number_of_neurons_per_core(sim.IF_curr_exp, 255)
# Définition des callbacks pour la récupération de l'écart-type sur les connexions entrée-intermédiaire, intermédiaire-sortie
weight_recorder1 = WeightRecorder(sampling_interval=1000.0,
projection=Conn_input_inter)
weight_recorder2 = WeightRecorder(sampling_interval=1000.0,
projection=Conn_inter_output)
simtime = ceil(max_time)
# Initialisation des tableaux pour la récupération des poids
weights_int = []
weights_out = []
try:
sim.run(simtime, callbacks=[weight_recorder1, weight_recorder2])
neo = Output.get_data(variables=["spikes", "v"])
spikes = neo.segments[0].spiketrains
#print(spikes)
v = neo.segments[0].filter(name='v')[0]
weights_int = Conn_input_inter.get(["weight"], format="list")
neo_in = Input.get_data(variables=["spikes"])
spikes_in = neo_in.segments[0].spiketrains
#print(spikes_in)
neo_intermediate = Intermediate.get_data(variables=["spikes", "v"])
spikes_intermediate = neo_intermediate.segments[0].spiketrains
#print(spikes_intermediate)
v_intermediate = neo_intermediate.segments[0].filter(name='v')[0]
#print(v_intermediate)
weights_out = Conn_inter_output.get(["weight"], format="list")
sim.reset()
sim.end()
except:
v = 0
spikes = 0
if (isinstance(spikes, list) and isinstance(v, AnalogSignal)):
# Récupération des écart-types
standard_deviation_out = weight_recorder2.get_standard_deviations()
standard_deviation_int = weight_recorder1.get_standard_deviations()
t = np.arange(0., max_time, 1.)
# Création et sauvegarde des graphs sur les spikes et écart-types
savePath = "./Generated_data/training/" + path + "/intermediate_layer_standard_deviation.png"
plt.plot(standard_deviation_int)
plt.xlabel("callbacks tick (1s)")
plt.ylabel("standard deviation of the weights( wmax=6, wmin=0.1 )")
plt.savefig(savePath)
plt.clf()
savePath = "./Generated_data/training/" + path + "/output_layer_standard_deviation.png"
plt.plot(standard_deviation_out)
plt.xlabel("callbacks tick")
plt.ylabel("standard deviation ( wmax=6, wmin=0.1 )")
plt.savefig(savePath)
plt.clf()
savePath = "./Generated_data/training/" + path + "/output_layer_membrane_voltage_and_spikes.png"
plot.Figure(
# plot voltage for first ([0]) neuron
plot.Panel(v,
ylabel="Membrane potential (mV)",
data_labels=[Output.label],
yticks=True,
xlim=(0, simtime)),
# plot spikes (or in this case spike)
plot.Panel(spikes, yticks=True, markersize=5, xlim=(0, simtime)),
title="Spiking activity of the output layer during training",
annotations="Simulated with {}".format(sim.name())).save(savePath)
savePath = "./Generated_data/training/" + path + "/intermediate_layer_membrane_voltage_and_spikes.png"
plot.Figure(
# plot voltage for first ([0]) neuron
plot.Panel(v_intermediate,
ylabel="Membrane potential (mV)",
data_labels=[Output.label],
yticks=True,
xlim=(0, simtime)),
# plot spikes (or in this case spike)
plot.Panel(spikes_intermediate,
yticks=True,
markersize=5,
xlim=(0, simtime)),
title="Spiking activity of the intermediate layer during training",
annotations="Simulated with {}".format(sim.name())).save(savePath)
return v, spikes, weights_int, weights_out
else:
print(
"simulation failed with parmaters parameters : (l'affichage des paramètres ayant causés le disfonctionnement de la simulation sera traitée à une date ultérieur, merci!)"
)
return 0, 0, 0, 0
