def test_weight(self):
proj = pynn.Projection(self.pop1, self.pop1, pynn.AllToAllConnector())
self.assertEqual(proj.get("weight", format="array"), 0)
pynn.run(None)
pynn.simulator.state.projections = []
synapse = pynn.standardmodels.synapses.StaticSynapse(weight=32)
proj = pynn.Projection(self.pop1, self.pop2, pynn.AllToAllConnector(),
synapse_type=synapse)
self.assertEqual(proj.get("weight", format="array"), 32)
proj.set(weight=42.5)
self.assertEqual(proj.get("weight", format="array"), 42)
proj.set(weight=43)
self.assertEqual(proj.get("weight", format="list"), [(0, 0, 43)])
with self.assertRaises(ValueError):
proj.set(weight=64)
with self.assertRaises(ValueError):
proj.set(weight=-1)
proj = pynn.Projection(self.pop3, self.pop4, pynn.AllToAllConnector(),
synapse_type=synapse)
connection_list = [(0, 0, 32), (1, 0, 32),
(0, 1, 32), (1, 1, 32),
(0, 2, 32), (1, 2, 32)]
self.assertEqual(proj.get("weight", format="list"), connection_list)
proj = pynn.Projection(self.pop4, self.pop3, pynn.AllToAllConnector(),
synapse_type=synapse)
connection_list = [(0, 0, 32), (1, 0, 32), (2, 0, 32),
(0, 1, 32), (1, 1, 32), (2, 1, 32)]
self.assertEqual(proj.get("weight", format="list"), connection_list)
pynn.run(None)
def test_bypass(self):
# Create network with single SpikeSourceArray which projects on a
# single HX neuron
spikes_in = self.poisson_spike_train(self.rate)
input_pop = pynn.Population(1,
pynn.cells.SpikeSourceArray,
cellparams={"spike_times": spikes_in})
output_pop = pynn.Population(1, pynn.cells.HXNeuron())
output_pop.record(["spikes"])
synapse = pynn.standardmodels.synapses.StaticSynapse(weight=63)
pynn.Projection(input_pop,
output_pop,
pynn.AllToAllConnector(),
synapse_type=synapse)
# Run experiment on chip
pynn.run(self.runtime)
# Make sure that at least 98% (estimate considering typical spike
# loss) of the spikes are read back and that the time difference
# between them is less than 0.01ms
spikes_out = output_pop.get_data().segments[0].spiketrains[0].magnitude
# Add a small number such that timestamps with a '5' the third decimal
# place are rounded up to the next higher (and not to the next even)
# number
input_set = set(np.round(spikes_in + 1e-9, 2))
output_set = set(np.round(spikes_out + 1e-9, 2))
assert len(input_set - output_set) / len(input_set) < 0.02
def test_number_of_spike_trains(self):
"""
Here we test that only spikes are recorded for the neurons we set in
the spike recording mode.
"""
runtime = 0.5 # ms
pop_size = 4
pops = []
pops.append(pynn.Population(pop_size, pynn.cells.HXNeuron()))
pops.append(pynn.Population(pop_size, pynn.cells.HXNeuron()))
pops.append(pynn.Population(pop_size, pynn.cells.HXNeuron()))
# Once record whole population, once subset and once no neurons
pops[0].record(['spikes'])
pynn.PopulationView(pops[1], [0]).record(['spikes'])
# Spike input
input_pop = pynn.Population(1, pynn.cells.SpikeSourceArray(
spike_times=np.arange(0.1, runtime, runtime / 10)))
for pop in pops:
pynn.Projection(input_pop, pop, pynn.AllToAllConnector(),
synapse_type=StaticSynapse(weight=63))
pynn.run(runtime)
# Assert correct number of recorded spike trains
self.assertEqual(len(pops[0].get_data().segments[0].spiketrains),
pop_size)
self.assertEqual(len(pops[1].get_data().segments[0].spiketrains), 1)
self.assertEqual(len(pops[2].get_data().segments[0].spiketrains), 0)
def main(params: dict):
pynn.setup()
nrn = pynn.Population(1, pynn.cells.HXNeuron(**params))
nrn.record(["spikes", "v"])
spike_times = [0.01, 0.03, 0.05, 0.07, 0.09, 0.11, 0.13, 0.15, 0.17, 0.19]
pop_input = pynn.Population(20,
pynn.cells.SpikeSourceArray,
cellparams={"spike_times": spike_times})
synapse = pynn.standardmodels.synapses.StaticSynapse(weight=63)
pynn.Projection(pop_input,
nrn,
pynn.AllToAllConnector(),
synapse_type=synapse)
pynn.run(0.2)
spiketrain = nrn.get_data("spikes").segments[0].spiketrains[0]
mem_v = nrn.get_data("v").segments[0]
membrane_times, membrane_voltage = zip(*mem_v.filter(name="v")[0])
pynn.end()
# Plot data
plt.figure()
plt.xlabel("Time [ms]")
plt.ylabel("Membrane Potential [LSB]")
plt.plot(membrane_times, membrane_voltage)
plt.savefig("plot_external_input.pdf")
plt.close()
return spiketrain
def test_injected_config():
pynn.setup(injected_config=pynn.InjectedConfiguration(
pre_non_realtime={halco.AtomicNeuronOnDLS(): lola.AtomicNeuron()}))
pynn.run(None)
pynn.end()
pynn.setup(injected_config=pynn.InjectedConfiguration(
post_non_realtime={halco.AtomicNeuronOnDLS(): lola.AtomicNeuron()
}))
pynn.run(None)
pynn.end()
pynn.setup(injected_config=pynn.InjectedConfiguration(
pre_realtime={halco.AtomicNeuronOnDLS(): lola.AtomicNeuron()}))
pynn.run(None)
pynn.end()
pynn.setup(injected_config=pynn.InjectedConfiguration(
post_realtime={halco.AtomicNeuronOnDLS(): lola.AtomicNeuron()}))
pynn.run(None)
pynn.end()
pynn.setup(injected_config=pynn.InjectedConfiguration(
pre_non_realtime={halco.AtomicNeuronOnDLS(): lola.AtomicNeuron()},
post_non_realtime={halco.AtomicNeuronOnDLS(): lola.AtomicNeuron()},
pre_realtime={halco.AtomicNeuronOnDLS(): lola.AtomicNeuron()},
post_realtime={halco.AtomicNeuronOnDLS(): lola.AtomicNeuron()}))
pynn.run(None)
pynn.end()
def test_number_of_neurons(self):
# TODO: grenade routing backend does not support 512 neurons?
for n_neurons in [1, 16, 32, 64, 128, 256]:
with self.subTest(n_neurons=n_neurons):
pynn.setup()
pynn.Population(n_neurons, pynn.cells.HXNeuron())
pynn.run(None)
pynn.end()
def test_delay(self):
proj = pynn.Projection(self.pop1, self.pop2, pynn.AllToAllConnector())
self.assertEqual(proj.get("delay", format="array"), [0])
proj.set(delay=0)
self.assertEqual(proj.get("delay", format="array"), [0])
with self.assertRaises(ValueError):
proj.set(delay=1)
pynn.run(None)
def main():
"""
Run a minimal pyNN-based experiment: Connect two neuron populations
and run an emulation. More elaborate examples of using BrainScaleS-2
through the pyNN-API can be found here:
https://github.com/electronicvisions/pynn-brainscales/tree/master/brainscales2/examples
"""
pynn.setup()
neurons_1 = get_neuron_population(2)
neurons_2 = get_neuron_population(3)
pynn.Projection(neurons_1, neurons_2, pynn.AllToAllConnector())
pynn.run(0.2)
pynn.end()
def test_timing_of_spikes(self):
"""
Test whether the output spikes have approximately the same timing as
the input spikes.
"""
runtime = 100 # ms
n_spikes = 1000
pop_1 = pynn.Population(1, pynn.cells.HXNeuron())
pop_1.record('spikes')
pop_2 = pynn.Population(1, pynn.cells.HXNeuron())
pop_2.record('spikes')
# Inject spikes
spikes_1 = np.arange(0.1, runtime, runtime / n_spikes)
input_pop_1 = pynn.Population(1, pynn.cells.SpikeSourceArray(
spike_times=spikes_1))
pynn.Projection(input_pop_1, pop_1, pynn.AllToAllConnector(),
synapse_type=StaticSynapse(weight=63))
# Take midpoints of spikes_1 to have some variation in timing and
spikes_2 = (spikes_1[:-1] + spikes_1[1:]) / 2
input_pop_2 = pynn.Population(1, pynn.cells.SpikeSourceArray(
spike_times=spikes_2))
pynn.Projection(input_pop_2, pop_2, pynn.AllToAllConnector(),
synapse_type=StaticSynapse(weight=63))
pynn.run(runtime)
# Make sure that at least 98% (estimate considering typical spike
# loss) of the spikes are read back and that the time difference
# between them is less than 0.01ms.
# To accomplish that we round the spike times to two decimal places
# and then compare the sets of the input and output spike times.
# By default numpy does not round '5' to the next higher number but
# to the next even number, e.g. np.round(2.5) == 2. Therefore, we
# add a small number such that timestamps with a '5' at the third
# decimal place are rounded up to the next higher (and not even) number
input_set_1 = set(np.round(spikes_1 + 1e-9, 2))
output_set_1 = set(np.round(
pop_1.get_data().segments[0].spiketrains[0].magnitude + 1e-9, 2))
self.assertLess(len(input_set_1 - output_set_1) / len(input_set_1),
0.02)
input_set_2 = set(np.round(spikes_2 + 1e-9, 2))
output_set_2 = set(np.round(
pop_2.get_data().segments[0].spiketrains[0].magnitude + 1e-9, 2))
self.assertLess(len(input_set_2 - output_set_2) / len(input_set_2),
0.02)
def get_isi(tau_ref: int):
pynn.setup()
cell_params.update({"refractory_period_refractory_time": tau_ref})
pop = pynn.Population(1, pynn.cells.HXNeuron(**cell_params))
pop.record("spikes")
pynn.run(0.2)
spikes = pop.get_data("spikes").segments[0].spiketrains[0]
if len(spikes) == 0:
return 0
isi = np.zeros(len(spikes) - 1)
for i in range(len(spikes) - 1):
isi[i] = spikes[i + 1] - spikes[i]
pynn.end()
return np.mean(isi)
def setUp(self):
self.bg_props = dict(
start=1, # ms
rate=20e3, # Hz
duration=100 # ms
)
# Emulate the network 1ms longer than Poisson stimulation, in order to
# convince oneself that the stimulation ends properly.
self.runtime = self.bg_props["start"] + self.bg_props["duration"] + 1
# The refractory time was found to must be set slightly larger 0 (DAC
# value) to achieve a short time on hardware (cf. #3741).
neuron_params = {"refractory_period_refractory_time": 5}
# By enabling the bypass mode, the neuron should spike at the income of
# each event.
pynn.setup(enable_neuron_bypass=True)
pop1 = pynn.Population(1, pynn.cells.HXNeuron(**neuron_params))
pop2 = pynn.Population(1, pynn.cells.HXNeuron(**neuron_params))
pop1.record(["spikes"])
pop2.record(["spikes"])
src = pynn.Population(1,
pynn.cells.SpikeSourcePoisson(**self.bg_props))
pynn.Projection(src,
pop1,
pynn.AllToAllConnector(),
synapse_type=StaticSynapse(weight=63))
pynn.Projection(src,
pop2,
pynn.AllToAllConnector(),
synapse_type=StaticSynapse(weight=63))
pynn.run(self.runtime)
self.spiketrain1 = pop1.get_data("spikes").segments[0].spiketrains[0]
self.spiketrain2 = pop2.get_data("spikes").segments[0].spiketrains[0]
pynn.reset()
self.rate3 = 10e3
src.set(rate=self.rate3)
pynn.run(self.runtime)
self.spiketrain3 = pop1.get_data("spikes").segments[1].spiketrains[0]
pynn.end()
def test_coco_extraction(self):
builder = sta.PlaybackProgramBuilderDumper()
an_coord0 = halco.AtomicNeuronOnDLS(halco.common.Enum(0))
an_coord1 = halco.AtomicNeuronOnDLS(halco.common.Enum(1))
neuron0 = lola.AtomicNeuron()
neuron0.leak.i_bias = 666
neuron1 = lola.AtomicNeuron()
neuron1.leak.i_bias = 420
builder.write(an_coord0, neuron0)
builder.write(an_coord1, neuron1)
common_config = hal.CommonNeuronBackendConfig()
common_config.clock_scale_fast = 3
common_coord = halco.CommonNeuronBackendConfigOnDLS()
builder.write(common_coord, common_config)
full_coco = {}
with tempfile.TemporaryDirectory() as tempdir:
filename = os.path.join(tempdir, "dump")
with open(filename, "wb") as fd:
fd.write(sta.to_binary(builder.done()))
full_coco = pynn.helper.coco_from_file(filename)
self.assertTrue(an_coord0 in full_coco)
self.assertTrue(an_coord1 in full_coco)
hx_coco = pynn.helper.filter_atomic_neuron(full_coco)
self.assertTrue(an_coord0 in hx_coco)
self.assertTrue(an_coord1 in hx_coco)
self.assertFalse(common_coord in hx_coco)
remainder_coco = pynn.helper.filter_non_atomic_neuron(full_coco)
self.assertFalse(an_coord0 in remainder_coco)
self.assertTrue(common_coord in remainder_coco)
self.assertEqual(remainder_coco[common_coord].clock_scale_fast, 3)
pynn.setup()
pop = pynn.Population(2, pynn.cells.HXNeuron(hx_coco))
self.assertTrue(numpy.array_equal(pop.get("leak_i_bias"), [666, 420]))
pynn.run(None)
pynn.end()
def main(params: dict):
pynn.setup()
nrn = pynn.Population(1, pynn.cells.HXNeuron(**params))
nrn.record(["spikes", "v"])
spike_times = [0.01, 0.03, 0.05, 0.07, 0.09, 0.11, 0.13, 0.15, 0.17, 0.19]
pop_input = pynn.Population(20, pynn.cells.SpikeSourceArray,
cellparams={"spike_times": spike_times})
synapse = pynn.standardmodels.synapses.StaticSynapse(weight=63)
pynn.Projection(pop_input, nrn, pynn.AllToAllConnector(),
synapse_type=synapse)
pynn.run(0.2)
spiketrain = nrn.get_data("spikes").segments[0].spiketrains[0]
mem_v = nrn.get_data("v").segments[0]
membrane_times, membrane_voltage = zip(*mem_v.filter(name="v")[0])
pynn.end()
return spiketrain, membrane_times, membrane_voltage
def main(params: dict):
log = pynn.logger.get("leak_over_threshold")
pynn.setup()
pop2 = pynn.Population(2, pynn.cells.HXNeuron(**params))
pop1 = pynn.Population(1, pynn.cells.HXNeuron(**params))
pop1.record(["spikes", "v"])
pop2.record("spikes")
pynn.run(0.2)
spikes1 = pop1.get_data("spikes").segments[0]
spikes2 = pop2.get_data("spikes").segments[0]
# TODO: Find out how to extract neuron ids.
for i, spiketrain in enumerate(spikes2.spiketrains):
log.INFO("Number of Spikes of Neuron {}: ".format(i + 1),
len(spiketrain))
log.INFO("Spiketimes of Neuron {}: ".format(i + 1), spiketrain)
spiketimes = spikes1.spiketrains[0]
log.INFO("Number of spikes of single recorded neuron: ", len(spiketimes))
log.INFO("Spiketimes of recorded neuron: ", spiketimes)
mem_v = pop1.get_data("v").segments[0]
times, membrane = zip(*mem_v.filter(name="v")[0])
log.INFO("Number of MADC Samples: ", len(times))
plt.figure()
plt.xlabel("Time [ms]")
plt.ylabel("Membrane Potential [LSB]")
plt.plot(times, membrane)
plt.savefig("plot_leak_over_threshold.pdf")
plt.close()
pynn.end()
# With this first projection, we connect the external spike source to the
# observed on-chip neuron population.
pynn.Projection(exc_stim_pop, stimulated_p,
pynn.AllToAllConnector(),
synapse_type=StaticSynapse(weight=63),
receptor_type="excitatory")
# Create off-chip populations serving as inhibitory external spike sources.
inh_spiketimes = [0.03]
inh_stim_pop = pynn.Population(1, SpikeSourceArray(spike_times=inh_spiketimes))
pynn.Projection(inh_stim_pop, stimulated_p,
pynn.AllToAllConnector(),
synapse_type=StaticSynapse(weight=63),
receptor_type="inhibitory")
'''
# You may play around with the parameters in this experiment to achieve
# different traces. Try to stack multiple PSPs, try to make the neurons spike,
# try to investigate differences between individual neuron instances,
# be creative!
pynn.run(0.1)
times, data = plot_membrane_dynamics(stimulated_p)
L.append(data)
pynn.end()
if i == 0 and j == 0:
np.save('times.npy', times)
LL.append(L)
plt.savefig('v' + str(i) + '.pdf')
np.save('LL.npy', np.array(LL))
plt.show()
def main():
main_log = pynn.logger.get("internal_projections_main")
pynn.setup()
lot_values = {
"threshold_v_threshold": 400,
"leak_v_leak": 1022,
"leak_i_bias": 420,
"leak_enable_division": True,
"reset_v_reset": 50,
"reset_i_bias": 950,
"reset_enable_multiplication": True,
"threshold_enable": True,
"membrane_capacitance_capacitance": 10,
"refractory_period_refractory_time": 95
}
cell_params = deepcopy(lot_values)
cell_params.update({
"leak_v_leak": 650,
"reset_v_reset": 650,
"excitatory_input_enable": True,
"excitatory_input_i_bias_tau": 150,
"excitatory_input_i_bias_gm": 200,
# FIXME: replace by i_drop_input and i_shift_reference
# "excitatory_input_v_syn": 700})
})
# not leak over threshold
pop1 = pynn.Population(1,
pynn.standardmodels.cells.HXNeuron(**cell_params))
# leak over threshold
pop2 = pynn.Population(100,
pynn.standardmodels.cells.HXNeuron(**lot_values))
pop1.record(["spikes", "v"])
pop2.record("spikes")
synapse = pynn.standardmodels.synapses.StaticSynapse(weight=63)
pynn.Projection(pop2, pop1, pynn.AllToAllConnector(), synapse_type=synapse)
pynn.run(0.2)
spiketimes1 = pop1.get_data("spikes").segments[0].spiketrains[0]
main_log.INFO("Spiketimes of first neuron: ", spiketimes1)
spikes2 = pop2.get_data("spikes").segments[0]
spikenumber2 = 0
for i in range(len(pop2)):
spikes = len(spikes2.spiketrains[i])
spikenumber2 += spikes
main_log.INFO("Number of spikes from pop2: ", spikenumber2)
mem_v = pop1.get_data("v").segments[0]
membrane_times, membrane_voltage = zip(*mem_v.filter(name="v")[0])
pynn.end()
# Plot data
plt.figure()
plt.xlabel("Time [ms]")
plt.ylabel("Membrane Potential [LSB]")
plt.plot(membrane_times, membrane_voltage)
plt.savefig("plot_internal_projections.pdf")
plt.close()
return len(spiketimes1)
def test_identical_connection(self):
pynn.Projection(self.pop1, self.pop2, pynn.AllToAllConnector())
pynn.Projection(self.pop1, self.pop2, pynn.AllToAllConnector())
pynn.run(None)
def test_empty():
pynn.setup()
pynn.run(0.2)
pynn.end()
def tearDown(self):
pynn.run(None)
pynn.end()
