def test_network_deterministic_delay(self):
dt = 0.01
G = 0.8
speed = 2.
sim = simulator.Simulator(
model=models.MontbrioPazoRoxin(),
coupling=coupling.Scaling(a=np.array([G])),
connectivity=self._random_network(speed=speed),
conduction_speed=speed,
monitors=[monitors.Raw()],
integrator=integrators.HeunStochastic(
dt=dt,
noise=noise.Additive(nsig=np.array([0.0, 0.0]),
noise_seed=42))).configure()
(pdq_t, pdq_d), = NbMPRBackend().run_sim(sim,
nstep=1,
compatibility_mode=True)
r_pdq, V_pdq = pdq_d[0, :, :, 0]
(raw_t, raw_d), = sim.run(simulation_length=1)
r_tvb, V_tvb = raw_d[0, :, :, 0]
np.testing.assert_allclose(r_tvb, r_pdq, rtol=1e-4)
np.testing.assert_allclose(V_tvb, V_pdq, rtol=1e-4)
def _get_run_sim(self, print_source=False):
template = '<%include file="nb-montbrio.py.mako"/>'
content = dict(foo='bar')
return NbMPRBackend().build_py_func(template,
content,
name='run_sim',
print_source=print_source)
def test_tavg_chunking(self):
dt = 0.01
G = 0.1
speed = 2.
sim = simulator.Simulator(
model=models.MontbrioPazoRoxin(),
coupling=coupling.Scaling(a=np.array([G])),
connectivity=self._random_network(speed=speed),
conduction_speed=speed,
monitors=[monitors.TemporalAverage(period=1)],
integrator=integrators.HeunStochastic(
dt=dt,
noise=noise.Additive(nsig=np.array([0.0, 0.0]),
noise_seed=42))).configure()
(pdq_t, pdq_d), = NbMPRBackend().run_sim(sim,
nstep=2000,
chunksize=2000,
compatibility_mode=True)
r_pdq = pdq_d[:, 0, :, 0]
V_pdq = pdq_d[:, 1, :, 0]
(pdq_chu_t,
pdq_chu_d), = NbMPRBackend().run_sim(sim,
nstep=2000,
chunksize=500,
compatibility_mode=True)
r_pdq_chu = pdq_d[:, 0, :, 0]
V_pdq_chu = pdq_d[:, 1, :, 0]
(raw_t, raw_d), = sim.run(simulation_length=20)
r_tvb = raw_d[:, 0, :, 0]
V_tvb = raw_d[:, 1, :, 0]
np.testing.assert_allclose(pdq_t, raw_t)
np.testing.assert_allclose(pdq_chu_t, raw_t)
np.testing.assert_allclose(r_pdq, r_tvb, rtol=1e-3)
np.testing.assert_allclose(V_pdq, V_tvb, rtol=1e-3)
# think a bit about the tolerances... TVB stores in floats, so that
# can accumulate. Might be a good idea to test agains history with
# double typed buffer.
np.testing.assert_allclose(r_pdq_chu, r_tvb, rtol=1e-3)
np.testing.assert_allclose(V_pdq_chu, V_tvb, rtol=1e-3)
def test_local_stochastic(self):
# run_sim = self._get_run_sim(print_source=True)
dt = 0.01
G = 0.
sim = simulator.Simulator(model=models.MontbrioPazoRoxin(),
coupling=coupling.Scaling(a=np.array([G])),
connectivity=self._random_network(),
conduction_speed=np.inf,
monitors=[monitors.Raw()],
integrator=integrators.HeunStochastic(
dt=dt,
noise=noise.Additive(
nsig=np.array([0.01, 0.02]),
noise_seed=42))).configure()
sim_det = simulator.Simulator(
model=models.MontbrioPazoRoxin(),
coupling=coupling.Scaling(a=np.array([G])),
connectivity=self._random_network(),
conduction_speed=np.inf,
monitors=[monitors.Raw()],
integrator=integrators.HeunDeterministic(dt=dt, )).configure()
(pdq_t, pdq_d), = NbMPRBackend().run_sim(sim, nstep=200)
(raw_t, raw_d), = sim.run(simulation_length=1)
(raw_t_det, raw_d_det), = sim_det.run(simulation_length=1)
r_tvb, V_tvb = raw_d[0, :, :, 0]
r_tvb_det, V_tvb_det = raw_d_det[0, :, :, 0]
r_pdq, V_pdq = pdq_d[0, :, :, 0]
np.testing.assert_allclose(
np.mean(r_tvb_det - r_tvb),
np.mean(r_tvb_det - r_pdq),
atol=1e-2 # zero mean doesn't play well with rtol
)
np.testing.assert_allclose(np.mean(V_tvb_det - V_tvb),
np.mean(V_tvb_det - V_pdq),
atol=1e-2)
np.testing.assert_allclose(np.std(r_tvb_det - r_tvb),
np.std(r_tvb_det - r_pdq),
rtol=1e-2)
np.testing.assert_allclose(np.std(V_tvb_det - V_tvb),
np.std(V_tvb_det - V_pdq),
rtol=1e-2)
def test_local_deterministic(self):
dt = 0.01
G = 0.
sim = simulator.Simulator(
model=models.MontbrioPazoRoxin(),
coupling=coupling.Scaling(a=np.array([G])),
connectivity=self._random_network(),
conduction_speed=np.inf,
monitors=[monitors.Raw()],
integrator=integrators.HeunStochastic( # matching integrator
dt=dt,
noise=noise.Additive(nsig=np.array([0.0, 0.0]),
noise_seed=42))).configure()
(pdq_t, pdq_d), = NbMPRBackend().run_sim(sim, nstep=1)
(raw_t, raw_d), = sim.run(simulation_length=1)
np.testing.assert_allclose(raw_d[0, :], pdq_d[0, :], rtol=1e-5)
def test_nb_pdq_100ms(benchmark):
sim = make_sim(100.0)
run_sim = NbMPRBackend().build_py_func(
'<%include file="nb-montbrio.py.mako"/>', {}, name='run_sim')
nstep = int(sim.simulation_length / sim.integrator.dt)
benchmark(lambda : run_sim(sim, nstep))