def parse_expressions(renderer, evaluator, numvalues=10):
exprs = [([m for m in get_identifiers(l) if len(m)==1], [], l.strip())
for l in TEST_EXPRESSIONS.split('\n') if l.strip()]
i, imod = 1, 33
for varids, funcids, expr in exprs:
pexpr = renderer.render_expr(expr)
n = 0
for _ in xrange(numvalues):
# assign some random values
ns = {}
for v in varids:
if v in ['n', 'm']: # integer values
ns[v] = i
else:
ns[v] = float(i)/imod
i = i%imod+1
r1 = eval(expr.replace('&', ' and ').replace('|', ' or '), ns)
n += 1
r2 = evaluator(pexpr, ns)
try:
# Use all close because we can introduce small numerical
# difference through sympy's rearrangements
assert_allclose(r1, r2, atol=10)
except AssertionError as e:
raise AssertionError("In expression " + str(expr) +
" translated to " + str(pexpr) +
" " + str(e))
def test_function_dependencies_numpy():
if prefs.codegen.target != 'numpy':
raise SkipTest('numpy-only test')
@implementation('cpp', '''
float foo(float x)
{
return 42*0.001;
}''')
@check_units(x=volt, result=volt)
def foo(x):
return 42*mV
# Second function with an independent implementation for C++ and an
# implementation for numpy that makes use of the previous function.
# Note that we don't need to use the explicit dependencies mechanism for
# numpy, since the Python function stores a reference to the referenced
# function directly
@implementation('cpp', '''
float bar(float x)
{
return 84*0.001;
}''')
@check_units(x=volt, result=volt)
def bar(x):
return 2*foo(x)
G = NeuronGroup(5, 'v : volt')
G.run_regularly('v = bar(v)')
net = Network(G)
net.run(defaultclock.dt)
assert_allclose(G.v_[:], 84*0.001)
def test_state_monitor_indexing():
# Check indexing semantics
G = NeuronGroup(10, 'v:volt')
G.v = np.arange(10) * volt
mon = StateMonitor(G, 'v', record=[5, 6, 7])
run(2 * defaultclock.dt)
assert_array_equal(mon.v, np.array([[5, 5],
[6, 6],
[7, 7]]) * volt)
assert_array_equal(mon.v_, np.array([[5, 5],
[6, 6],
[7, 7]]))
assert_array_equal(mon[5].v, mon.v[0])
assert_array_equal(mon[7].v, mon.v[2])
assert_array_equal(mon[[5, 7]].v, mon.v[[0, 2]])
assert_array_equal(mon[np.array([5, 7])].v, mon.v[[0, 2]])
assert_allclose(mon.t[1:], Quantity([defaultclock.dt]))
assert_raises(IndexError, lambda: mon[8])
assert_raises(TypeError, lambda: mon['string'])
assert_raises(TypeError, lambda: mon[5.0])
assert_raises(TypeError, lambda: mon[[5.0, 6.0]])
def test_function_dependencies_cython():
if prefs.codegen.target != 'cython':
raise SkipTest('cython-only test')
@implementation('cython', '''
cdef float foo(float x):
return 42*0.001
''')
@check_units(x=volt, result=volt)
def foo(x):
return 42*mV
# Second function with an independent implementation for numpy and an
# implementation for C++ that makes use of the previous function.
@implementation('cython', '''
cdef float bar(float x):
return 2*foo(x)
''', dependencies={'foo': foo})
@check_units(x=volt, result=volt)
def bar(x):
return 84*mV
G = NeuronGroup(5, 'v : volt')
G.run_regularly('v = bar(v)')
net = Network(G)
net.run(defaultclock.dt)
assert_allclose(G.v_[:], 84*0.001)
def test_inplace_on_scalars():
# We want "copy semantics" for in-place operations on scalar quantities
# in the same way as for Python scalars
for scalar in [3*mV, 3*mV/mV]:
scalar_reference = scalar
scalar_copy = Quantity(scalar, copy=True)
scalar += scalar_copy
assert_equal(scalar_copy, scalar_reference)
scalar *= 1.5
assert_equal(scalar_copy, scalar_reference)
scalar /= 2
assert_equal(scalar_copy, scalar_reference)
# also check that it worked correctly for the scalar itself
assert_allclose(scalar, (scalar_copy + scalar_copy)*1.5/2)
# For arrays, it should use reference semantics
for vector in [[3]*mV, [3]*mV/mV]:
vector_reference = vector
vector_copy = Quantity(vector, copy=True)
vector += vector_copy
assert_equal(vector, vector_reference)
vector *= 1.5
assert_equal(vector, vector_reference)
vector /= 2
assert_equal(vector, vector_reference)
# also check that it worked correctly for the vector itself
assert_allclose(vector, (vector_copy + vector_copy)*1.5/2)
def test_math_functions():
'''
Test that math functions give the same result, regardless of whether used
directly or in generated Python or C++ code.
'''
default_dt = defaultclock.dt
test_array = np.array([-1, -0.5, 0, 0.5, 1])
def int_(x):
return array(x, dtype=int)
int_.__name__ = 'int'
with catch_logs() as _: # Let's suppress warnings about illegal values
# Functions with a single argument
for func in [cos, tan, sinh, cosh, tanh,
arcsin, arccos, arctan,
log, log10,
exp, np.sqrt,
np.ceil, np.floor, np.sign, int_]:
# Calculate the result directly
numpy_result = func(test_array)
# Calculate the result in a somewhat complicated way by using a
# subexpression in a NeuronGroup
if func.__name__ == 'absolute':
# we want to use the name abs instead of absolute
func_name = 'abs'
else:
func_name = func.__name__
G = NeuronGroup(len(test_array),
'''func = {func}(variable) : 1
variable : 1'''.format(func=func_name))
G.variable = test_array
mon = StateMonitor(G, 'func', record=True)
net = Network(G, mon)
net.run(default_dt)
assert_allclose(numpy_result, mon.func_.flatten(),
err_msg='Function %s did not return the correct values' % func.__name__)
# Functions/operators
scalar = 3
for func, operator in [(np.power, '**'), (np.mod, '%')]:
# Calculate the result directly
numpy_result = func(test_array, scalar)
# Calculate the result in a somewhat complicated way by using a
# subexpression in a NeuronGroup
G = NeuronGroup(len(test_array),
'''func = variable {op} scalar : 1
variable : 1'''.format(op=operator))
G.variable = test_array
mon = StateMonitor(G, 'func', record=True)
net = Network(G, mon)
net.run(default_dt)
assert_allclose(numpy_result, mon.func_.flatten(),
err_msg='Function %s did not return the correct values' % func.__name__)
def test_rate_monitor_2():
G = NeuronGroup(10, 'v : 1', threshold='v>1') # no reset
G.v['i<5'] = 1.1 # Half of the neurons fire every time step
rate_mon = PopulationRateMonitor(G)
net = Network(G, rate_mon)
net.run(10*defaultclock.dt)
assert_allclose(rate_mon.rate, 0.5 * np.ones(10) / defaultclock.dt)
assert_allclose(rate_mon.rate_, 0.5 *np.asarray(np.ones(10) / defaultclock.dt_))
def test_timedarray_no_upsampling():
# Test a TimedArray where no upsampling is necessary because the monitor's
# dt is bigger than the TimedArray's
ta = TimedArray(np.arange(10), dt=0.01*ms)
G = NeuronGroup(1, 'value = ta(t): 1', dt=0.1*ms)
mon = StateMonitor(G, 'value', record=True, dt=1*ms)
run(2.1*ms)
assert_allclose(mon[0].value, [0, 9, 9])
def assert_quantity(q, values, unit):
assert isinstance(q, Quantity) or (have_same_dimensions(unit, 1) and
(values.shape == () or
isinstance(q, np.ndarray))), q
assert_allclose(np.asarray(q), values)
assert have_same_dimensions(q, unit), ('Dimension mismatch: (%s) (%s)' %
(get_dimensions(q),
get_dimensions(unit)))
def test_timedarray_semantics():
# Make sure that timed arrays are interpreted as specifying the values
# between t and t+dt (not between t-dt/2 and t+dt/2 as in Brian1)
ta = TimedArray(array([0, 1]), dt=0.4*ms)
G = NeuronGroup(1, 'value = ta(t) : 1', dt=0.1*ms)
mon = StateMonitor(G, 'value', record=0)
run(0.8*ms)
assert_allclose(mon[0].value, [0, 0, 0, 0, 1, 1, 1, 1])
assert_allclose(mon[0].value, ta(mon.t))
def test_rallpack1():
'''
Rallpack 1
'''
if prefs.core.default_float_dtype is np.float32:
raise SkipTest('Need double precision for this test')
defaultclock.dt = 0.05*ms
# Morphology
diameter = 1*um
length = 1*mm
Cm = 1 * uF / cm ** 2
Ri = 100 * ohm * cm
N = 1000
morpho = Cylinder(diameter=diameter, length=length, n=N)
# Passive channels
gL = 1./(40000*ohm*cm**2)
EL = -65*mV
eqs = '''
Im = gL*(EL - v) : amp/meter**2
I : amp (point current, constant)
'''
neuron = SpatialNeuron(morphology=morpho, model=eqs, Cm=Cm, Ri=Ri)
neuron.v = EL
neuron.I[0] = 0.1*nA # injecting at the left end
#Record at the two ends
mon = StateMonitor(neuron, 'v', record=[0, 999], when='start', dt=0.05*ms)
run(250*ms + defaultclock.dt)
# Load the theoretical results
basedir = os.path.join(os.path.abspath(os.path.dirname(__file__)),
'rallpack_data')
data_0 = np.loadtxt(os.path.join(basedir, 'ref_cable.0'))
data_x = np.loadtxt(os.path.join(basedir, 'ref_cable.x'))
scale_0 = max(data_0[:, 1]*volt) - min(data_0[:, 1]*volt)
scale_x = max(data_x[:, 1]*volt) - min(data_x[:, 1]*volt)
squared_diff_0 = (data_0[:, 1] * volt - mon[0].v)**2
squared_diff_x = (data_x[:, 1] * volt - mon[999].v)**2
rel_RMS_0 = sqrt(mean(squared_diff_0))/scale_0
rel_RMS_x = sqrt(mean(squared_diff_x))/scale_x
max_rel_0 = sqrt(max(squared_diff_0))/scale_0
max_rel_x = sqrt(max(squared_diff_x))/scale_x
# sanity check: times are the same
assert_allclose(mon.t/second, data_0[:, 0])
assert_allclose(mon.t/second, data_x[:, 0])
# RMS error should be < 0.1%, maximum error along the curve should be < 0.5%
assert 100*rel_RMS_0 < 0.1
assert 100*rel_RMS_x < 0.1
assert 100*max_rel_0 < 0.5
assert 100*max_rel_x < 0.5
def test_constants_values():
'''
Make sure that symbolic constants use the correct values in code
'''
G = NeuronGroup(3, 'v : 1')
G.v[0] = 'pi'
G.v[1] = 'e'
G.v[2] = 'inf'
run(0*ms)
assert_allclose(G.v[:], [np.pi, np.e, np.inf])
def test_poissoninput_refractory():
eqs = '''
dv/dt = 0/second : 1 (unless refractory)
'''
G = NeuronGroup(10, eqs, reset='v=0', threshold='v>4.5', refractory=5*defaultclock.dt)
# Will increase the value by 1.0 at each time step
P = PoissonInput(G, 'v', 1, 1/defaultclock.dt, weight=1.0)
mon = StateMonitor(G, 'v', record=5)
run(10*defaultclock.dt)
expected = np.arange(10, dtype=np.float)
expected[6-int(schedule_propagation_offset()/defaultclock.dt):] = 0
assert_allclose(mon[5].v[:], expected)
def test_active_flag():
G = NeuronGroup(1, 'dv/dt = 1/ms : 1')
mon = StateMonitor(G, 'v', record=0)
mon.active = False
run(1*ms)
mon.active = True
G.active = False
run(1*ms)
device.build(direct_call=False, **device.build_options)
# Monitor should start recording at 1ms
# Neurongroup should not integrate after 1ms (but should have integrated before)
assert_allclose(mon[0].t[0], 1*ms)
assert_allclose(mon[0].v, 1.0)
def test_str_repr():
'''
Test that str representations do not raise any errors and that repr
fullfills eval(repr(x)) == x.
'''
from numpy import array # necessary for evaluating repr
units_which_should_exist = [metre, meter, kilogram, kilogramme, second, amp, kelvin, mole, candle,
radian, steradian, hertz, newton, pascal, joule, watt,
coulomb, volt, farad, ohm, siemens, weber, tesla, henry,
lumen, lux, becquerel, gray, sievert, katal,
gram, gramme, molar, liter, litre]
# scaled versions of all these units should exist (we just check farad as an example)
some_scaled_units = [Yfarad, Zfarad, Efarad, Pfarad, Tfarad, Gfarad, Mfarad, kfarad,
hfarad, dafarad, dfarad, cfarad, mfarad, ufarad, nfarad, pfarad,
ffarad, afarad, zfarad, yfarad]
# some powered units
powered_units = [cmetre2, Yfarad3]
# Combined units
complex_units = [(kgram * metre2)/(amp * second3),
5 * (kgram * metre2)/(amp * second3),
metre * second**-1, 10 * metre * second**-1,
array([1, 2, 3]) * kmetre / second,
np.ones(3) * nS / cm**2,
Unit(1, dim=get_or_create_dimension(length=5, time=2)),
8000*umetre**3, [0.0001, 10000] * umetre**3,
1/metre, 1/(coulomb*metre**2), Unit(1)/second,
3.*mM, 5*mole/liter, 7*liter/meter3]
unitless = [second/second, 5 * second/second, Unit(1)]
for u in itertools.chain(units_which_should_exist, some_scaled_units,
powered_units, complex_units, unitless):
assert(len(str(u)) > 0)
assert_allclose(eval(repr(u)), u)
# test the `DIMENSIONLESS` object
assert str(DIMENSIONLESS) == '1'
assert repr(DIMENSIONLESS) == 'Dimension()'
# test DimensionMismatchError (only that it works without raising an error
for error in [DimensionMismatchError('A description'),
DimensionMismatchError('A description', DIMENSIONLESS),
DimensionMismatchError('A description', DIMENSIONLESS,
second.dim)]:
assert len(str(error))
assert len(repr(error))
def test_state_monitor_record_single_timestep():
G = NeuronGroup(1, 'dv/dt = -v/(5*ms) : 1')
G.v = 1
mon = StateMonitor(G, 'v', record=True)
# Recording before a run should not work
assert_raises(TypeError, lambda: mon.record_single_timestep())
run(0.5*ms)
mon.record_single_timestep()
device.build(direct_call=False, **device.build_options)
assert mon.t[0] == 0*ms
assert mon[0].v[0] == 1
assert_allclose(mon.t[-1], 0.5*ms)
assert len(mon.t) == 6
assert mon[0].v[-1] == G.v
def test_spikegenerator_period():
'''
Basic test for `SpikeGeneratorGroup`.
'''
indices = np.array([3, 2, 1, 1, 2, 3, 3, 2, 1])
times = np.array([1, 4, 4, 3, 2, 4, 2, 3, 2]) * ms
SG = SpikeGeneratorGroup(5, indices, times, period=5*ms)
s_mon = SpikeMonitor(SG)
run(10*ms)
for idx in xrange(5):
generator_spikes = sorted([(idx, time) for time in times[indices==idx]] + [(idx, time+5*ms) for time in times[indices==idx]])
recorded_spikes = sorted([(idx, time) for time in s_mon.t[s_mon.i==idx]])
assert_allclose(generator_spikes, recorded_spikes)
def test_spikegenerator_extreme_period():
'''
Basic test for `SpikeGeneratorGroup`.
'''
indices = np.array([0, 1, 2])
times = np.array([0, 1, 2]) * ms
SG = SpikeGeneratorGroup(5, indices, times, period=1e6*second)
s_mon = SpikeMonitor(SG)
with catch_logs() as l:
run(10*ms)
assert_equal(s_mon.i, np.array([0, 1, 2]))
assert_allclose(s_mon.t, [0, 1, 2]*ms)
assert len(l) == 1 and l[0][1].endswith('spikegenerator_long_period')
def test_long_timedarray():
'''
Use a very long timedarray (with a big dt), where the upsampling can lead
to integer overflow.
'''
ta = TimedArray(np.arange(16385), dt=1*second)
G = NeuronGroup(1, 'value = ta(t) : 1')
mon = StateMonitor(G, 'value', record=True)
net = Network(G, mon)
# We'll start the simulation close to the critical boundary
# FIXME: setting the time like this does not work for standalone
net.t_ = float(16384*second - 5*ms)
net.run(10*ms)
assert_allclose(mon[0].value[mon.t < 16384*second], 16383)
assert_allclose(mon[0].value[mon.t >= 16384*second], 16384)
def test_state_variables_string_indices():
'''
Test accessing subgroups with string indices.
'''
G = NeuronGroup(10, 'v : volt')
SG = G[4:9]
assert len(SG.v['i>3']) == 1
G.v = np.arange(10) * mV
assert len(SG.v['v>7.5*mV']) == 1
# Combined string indexing and assignment
SG.v['i > 3'] = 'i*10*mV'
assert_allclose(G.v[:], [0, 1, 2, 3, 4, 5, 6, 7, 40, 9] * mV)
def test_rate_monitor_1():
G = NeuronGroup(5, 'v : 1', threshold='v>1') # no reset
G.v = 1.1 # All neurons spike every time step
rate_mon = PopulationRateMonitor(G)
run(10*defaultclock.dt)
assert_allclose(rate_mon.t, np.arange(10) * defaultclock.dt)
assert_allclose(rate_mon.t_, np.arange(10) * defaultclock.dt_)
assert_allclose(rate_mon.rate, np.ones(10) / defaultclock.dt)
assert_allclose(rate_mon.rate_, np.asarray(np.ones(10) / defaultclock.dt_))
def test_timedarray_direct_use():
ta = TimedArray(np.linspace(0, 10, 11), 1*ms)
assert ta(-1*ms) == 0
assert ta(5*ms) == 5
assert ta(10*ms) == 10
assert ta(15*ms) == 10
ta = TimedArray(np.linspace(0, 10, 11)*amp, 1*ms)
assert ta(-1*ms) == 0*amp
assert ta(5*ms) == 5*amp
assert ta(10*ms) == 10*amp
assert ta(15*ms) == 10*amp
ta2d = TimedArray((np.linspace(0, 11, 12)*amp).reshape(4, 3), 1*ms)
assert ta2d(-1*ms, 0) == 0*amp
assert ta2d(0*ms, 0) == 0*amp
assert ta2d(0*ms, 1) == 1*amp
assert ta2d(1*ms, 1) == 4*amp
assert_allclose(ta2d(1*ms, [0, 1, 2]), [3, 4, 5]*amp)
assert_allclose(ta2d(15*ms, [0, 1, 2]), [9, 10, 11]*amp)
def test_synapses_state_monitor():
G = NeuronGroup(2, '')
S = Synapses(G, G, 'w: siemens')
S.connect(True)
S.w = 'j*nS'
# record from a Synapses object (all synapses connecting to neuron 1)
synapse_mon = StateMonitor(S, 'w', record=S[:, 1])
synapse_mon2 = StateMonitor(S, 'w', record=S['j==1'])
net = Network(G, S, synapse_mon, synapse_mon2)
net.run(10*ms)
# Synaptic variables
assert_allclose(synapse_mon[S[0, 1]].w, 1*nS)
assert_allclose(synapse_mon.w[1], 1*nS)
assert_allclose(synapse_mon2[S[0, 1]].w, 1*nS)
assert_allclose(synapse_mon2[S['i==0 and j==1']].w, 1*nS)
assert_allclose(synapse_mon2.w[1], 1*nS)
def test_custom_events():
# Set (could be moved in a setup)
EL = -65*mV
gL = 0.0003*siemens/cm**2
ev = '''
Im = gL * (EL - v) : amp/meter**2
event_time1 : second
'''
# Create a three compartments morphology
morpho = Soma(diameter=10*um)
morpho.dend1 = Cylinder(n=1, diameter=1*um, length=10*um )
morpho.dend2 = Cylinder(n=1, diameter=1*um, length=10*um )
G = SpatialNeuron(morphology=morpho,
model=ev,
events={'event1': 't>=i*ms and t<i*ms+dt'})
G.run_on_event('event1', 'event_time1 = 0.1*ms')
run(0.2*ms)
# Event has size three now because there are three compartments
assert_allclose(G.event_time1[:], [0.1, 0, 0]*ms)
def test_constants():
import brian2.units.constants as constants
# Check that the expected names exist and have the correct dimensions
assert constants.avogadro_constant.dim == (1/mole).dim
assert constants.boltzmann_constant.dim == (joule/kelvin).dim
assert constants.electric_constant.dim == (farad/meter).dim
assert constants.electron_mass.dim == kilogram.dim
assert constants.elementary_charge.dim == coulomb.dim
assert constants.faraday_constant.dim == (coulomb/mole).dim
assert constants.gas_constant.dim == (joule/mole/kelvin).dim
assert constants.magnetic_constant.dim == (newton/amp2).dim
assert constants.molar_mass_constant.dim == (kilogram/mole).dim
assert constants.zero_celsius.dim == kelvin.dim
# Check the consistency between a few constants
assert_allclose(constants.gas_constant,
constants.avogadro_constant*constants.boltzmann_constant)
assert_allclose(constants.faraday_constant,
constants.avogadro_constant*constants.elementary_charge)
def test_rate_monitor_subgroups_2():
G = NeuronGroup(6, '''do_spike : boolean''', threshold='do_spike')
G.do_spike = [True, False, False, False, True, True]
rate_all = PopulationRateMonitor(G)
rate_1 = PopulationRateMonitor(G[:2])
rate_2 = PopulationRateMonitor(G[2:4])
rate_3 = PopulationRateMonitor(G[4:])
run(2*defaultclock.dt)
assert_allclose(rate_all.rate, 0.5 / defaultclock.dt)
assert_allclose(rate_1.rate, 0.5 / defaultclock.dt)
assert_allclose(rate_2.rate, 0*Hz)
assert_allclose(rate_3.rate, 1 / defaultclock.dt)
def test_spatialneuron_capacitive_currents():
if prefs.core.default_float_dtype is np.float32:
raise SkipTest('Need double precision for this test')
defaultclock.dt = 0.1*ms
morpho = Cylinder(x=[0, 10]*cm, diameter=2*238*um, n=200, type='axon')
El = 10.613* mV
ENa = 115*mV
EK = -12*mV
gl = 0.3*msiemens/cm**2
gNa0 = 120*msiemens/cm**2
gK = 36*msiemens/cm**2
# Typical equations
eqs = '''
# The same equations for the whole neuron, but possibly different parameter values
# distributed transmembrane current
Im = gl * (El-v) + gNa * m**3 * h * (ENa-v) + gK * n**4 * (EK-v) : amp/meter**2
I : amp (point current) # applied current
dm/dt = alpham * (1-m) - betam * m : 1
dn/dt = alphan * (1-n) - betan * n : 1
dh/dt = alphah * (1-h) - betah * h : 1
alpham = (0.1/mV) * (-v+25*mV) / (exp((-v+25*mV) / (10*mV)) - 1)/ms : Hz
betam = 4 * exp(-v/(18*mV))/ms : Hz
alphah = 0.07 * exp(-v/(20*mV))/ms : Hz
betah = 1/(exp((-v+30*mV) / (10*mV)) + 1)/ms : Hz
alphan = (0.01/mV) * (-v+10*mV) / (exp((-v+10*mV) / (10*mV)) - 1)/ms : Hz
betan = 0.125*exp(-v/(80*mV))/ms : Hz
gNa : siemens/meter**2
'''
neuron = SpatialNeuron(morphology=morpho, model=eqs, Cm=1*uF/cm**2,
Ri=35.4*ohm*cm, method="exponential_euler")
mon = StateMonitor(neuron, ['Im', 'Ic'], record=True, when='end')
run(10*ms)
neuron.I[0] = 1*uA # current injection at one end
run(3*ms)
neuron.I = 0*amp
run(10*ms)
device.build(direct_call=False, **device.build_options)
assert_allclose((mon.Im-mon.Ic).sum(axis=0)/(mA/cm**2), np.zeros(230),
atol=1e6)
def test_infinitecable():
'''
Test simulation of an infinite cable vs. theory for current pulse (Green function)
'''
BrianLogger.suppress_name('resolution_conflict')
defaultclock.dt = 0.001*ms
# Morphology
diameter = 1*um
Cm = 1 * uF / cm ** 2
Ri = 100 * ohm * cm
N = 500
morpho=Cylinder(diameter=diameter,length=3*mm,n=N)
# Passive channels
gL=1e-4*siemens/cm**2
eqs='''
Im=-gL*v : amp/meter**2
I : amp (point current)
'''
neuron = SpatialNeuron(morphology=morpho, model=eqs, Cm=Cm, Ri=Ri)
# Monitors
mon=StateMonitor(neuron,'v',record=N/2-20)
neuron.I[len(neuron)//2]=1*nA # injecting in the middle
run(0.02*ms)
neuron.I=0*amp
run(3*ms)
t = mon.t
v = mon[N//2-20].v
# Theory (incorrect near cable ends)
x = 20*morpho.length[0]
la = neuron.space_constant[0]
taum = Cm/gL # membrane time constant
theory = 1./(la*Cm*pi*diameter)*sqrt(taum/(4*pi*(t+defaultclock.dt)))*\
exp(-(t+defaultclock.dt)/taum-taum/(4*(t+defaultclock.dt))*(x/la)**2)
theory = theory*1*nA*0.02*ms
assert_allclose(v[t>0.5*ms],theory[t>0.5*ms], rtol=1e14, atol=1e10) # high error tolerance (not exact because not infinite cable)
def test_finitecable():
'''
Test simulation of short cylinder vs. theory for constant current.
'''
if prefs.core.default_float_dtype is np.float32:
raise SkipTest('Need double precision for this test')
BrianLogger.suppress_name('resolution_conflict')
defaultclock.dt = 0.01*ms
# Morphology
diameter = 1*um
length = 300*um
Cm = 1 * uF / cm ** 2
Ri = 150 * ohm * cm
N = 200
morpho=Cylinder(diameter=diameter,length=length,n=N)
# Passive channels
gL=1e-4*siemens/cm**2
EL=-70*mV
eqs='''
Im=gL*(EL-v) : amp/meter**2
I : amp (point current)
'''
neuron = SpatialNeuron(morphology=morpho, model=eqs, Cm=Cm, Ri=Ri)
neuron.v = EL
neuron.I[0]=0.02*nA # injecting at the left end
run(100*ms)
# Theory
x = neuron.distance
v = neuron.v
la = neuron.space_constant[0]
ra = la*4*Ri/(pi*diameter**2)
theory = EL+ra*neuron.I[0]*cosh((length-x)/la)/sinh(length/la)
assert_allclose(v-EL, theory-EL, rtol=1e12, atol=1e8)
def test_clip():
G = NeuronGroup(4, '''
clipexpr1 = clip(integer_var1, 0, 1) : integer
clipexpr2 = clip(integer_var2, -0.5, 1.5) : integer
clipexpr3 = clip(float_var1, 0, 1) : 1
clipexpr4 = clip(float_var2, -0.5, 1.5) : 1
integer_var1 : integer
integer_var2 : integer
float_var1 : 1
float_var2 : 1
''')
G.integer_var1 = [0, 1, -1, 2]
G.integer_var2 = [0, 1, -1, 2]
G.float_var1 = [0., 1., -1., 2.]
G.float_var2 = [0., 1., -1., 2.]
s_mon = StateMonitor(G, ['clipexpr1', 'clipexpr2',
'clipexpr3', 'clipexpr4'], record=True)
run(defaultclock.dt)
assert_equal(s_mon.clipexpr1.flatten(), [0, 1, 0, 1])
assert_equal(s_mon.clipexpr2.flatten(), [0, 1, 0, 1])
assert_allclose(s_mon.clipexpr3.flatten(), [0, 1, 0, 1])
assert_allclose(s_mon.clipexpr4.flatten(), [0, 1, -0.5, 1.5])
def test_conditional_write_automatic_and_manual():
source = NeuronGroup(1, '', threshold='True') # spiking all the time
target = NeuronGroup(
2,
'''dv/dt = 0/ms : 1 (unless refractory)
dw/dt = 0/ms : 1''',
threshold='t == 0*ms',
refractory='False') # only refractory for the first time step
# Cell is spiking/refractory only in the first time step
syn = Synapses(source,
target,
on_pre='''v += 1
w += 1 * int(not_refractory_post)'''
)
syn.connect()
mon = StateMonitor(target, ['v', 'w'], record=True, when='end')
run(2 * defaultclock.dt)
# Synapse should not have been effective in the first time step
assert_allclose(mon.v[:, 0], 0)
assert_allclose(mon.v[:, 1], 1)
assert_allclose(mon.w[:, 0], 0)
assert_allclose(mon.w[:, 1], 1)
def test_rate_monitor_subgroups():
old_dt = defaultclock.dt
defaultclock.dt = 0.01 * ms
G = NeuronGroup(4,
'''dv/dt = rate : 1
rate : Hz''',
threshold='v>0.999',
reset='v=0')
G.rate = [100, 200, 400, 800] * Hz
rate_all = PopulationRateMonitor(G)
rate_1 = PopulationRateMonitor(G[:2])
rate_2 = PopulationRateMonitor(G[2:])
run(1 * second)
assert_allclose(mean(G.rate_[:]), mean(rate_all.rate_[:]))
assert_allclose(mean(G.rate_[:2]), mean(rate_1.rate_[:]))
assert_allclose(mean(G.rate_[2:]), mean(rate_2.rate_[:]))
defaultclock.dt = old_dt
def test_refractoriness_variables():
# Try a string evaluating to a quantity an an explicit boolean
# condition -- all should do the same thing
for ref_time in ['5*ms', '(t-lastspike + 1e-3*dt) < 5*ms',
'time_since_spike + 1e-3*dt < 5*ms', 'ref_subexpression',
'(t-lastspike + 1e-3*dt) <= ref', 'ref', 'ref_no_unit*ms']:
reinit_and_delete()
G = NeuronGroup(1, '''
dv/dt = 99.999*Hz : 1 (unless refractory)
dw/dt = 99.999*Hz : 1
ref : second
ref_no_unit : 1
time_since_spike = (t - lastspike) +1e-3*dt : second
ref_subexpression = (t - lastspike + 1e-3*dt) < ref : boolean
''',
threshold='v>1', reset='v=0;w=0',
refractory=ref_time,
dtype={'ref': defaultclock.variables['t'].dtype,
'ref_no_unit': defaultclock.variables['t'].dtype,
'lastspike': defaultclock.variables['t'].dtype,
'time_since_spike': defaultclock.variables['t'].dtype})
G.ref = 5*ms
G.ref_no_unit = 5
# It should take 10ms to reach the threshold, then v should stay at 0
# for 5ms, while w continues to increase
mon = StateMonitor(G, ['v', 'w'], record=True, when='end')
run(20*ms)
try:
# No difference before the spike
assert_allclose(mon[0].v[:timestep(10*ms, defaultclock.dt)],
mon[0].w[:timestep(10*ms, defaultclock.dt)])
# v is not updated during refractoriness
in_refractoriness = mon[0].v[timestep(10*ms, defaultclock.dt):timestep(15*ms, defaultclock.dt)]
assert_allclose(in_refractoriness, np.zeros_like(in_refractoriness))
# w should evolve as before
assert_allclose(mon[0].w[:timestep(5*ms, defaultclock.dt)],
mon[0].w[timestep(10*ms, defaultclock.dt)+1:timestep(15*ms, defaultclock.dt)+1])
assert np.all(mon[0].w[timestep(10*ms, defaultclock.dt)+1:timestep(15*ms, defaultclock.dt)+1] > 0)
# After refractoriness, v should increase again
assert np.all(mon[0].v[timestep(15*ms, defaultclock.dt):timestep(20*ms, defaultclock.dt)] > 0)
except AssertionError as ex:
raise
raise AssertionError('Assertion failed when using %r as refractory argument:\n%s' % (ref_time, ex))
def test_timedarray_no_units():
ta = TimedArray(np.arange(10), dt=0.1 * ms)
G = NeuronGroup(1, 'value = ta(t) + 1: 1', dt=0.1 * ms)
mon = StateMonitor(G, 'value', record=True, dt=0.1 * ms)
run(1.1 * ms)
assert_allclose(mon[0].value_, np.clip(np.arange(len(mon[0].t)), 0, 9) + 1)
def test_CudaSpikeQueue_push_outer_loop2():
# This test was testing a special case scenario in an old version of
# CudaSpikeQueue::push(). It is now left here as a sanity check for spike
# propagation.
# This test creates a situation where one pre neuron is connected to 2048
# post neurons, where all synapses have different delays except of the the
# two synapses with ids 1023 and 1024 (last thread of first loop cycle and
# first thread of last loop cycle in CudaSpikeQueue::push()), which have
# the same delay. Additionaly all delay queues have different size when the
# pre neuron spikes one time. The different delay queue sizes is just a
# leftover of another test and does not have any effect on this test, but
# I'll leave it as an example.
num_blocks = 1
prefs['devices.cuda_standalone.parallel_blocks'] = num_blocks
# make sure we don't use less then 1024 threads due to register usage
prefs['devices.cuda_standalone.cuda_backend.extra_compile_args_nvcc'] += [
'-maxrregcount=64'
]
threads_per_block = 1024
neurons_per_block = 2 * threads_per_block
N = neurons_per_block * num_blocks # each block gets 2048 synapses to push
default_dt = defaultclock.dt
# for the first connection each thread has a different delay per block of
# postsynatpic neurons
delays_one_block = arange(neurons_per_block) * default_dt
delays0 = tile(delays_one_block, num_blocks)
# for the second connection we want to trigger situation (a), so per block the
# last thread of first loop and first thread of last loop have the same delay
delays_one_block[threads_per_block] = delays_one_block[threads_per_block -
1]
delays = tile(delays_one_block, num_blocks)
# inp neuron 0 fires the first (2*threads_per_block) time steps, s.t. all queues
# have different size after that (first 2048, last 0)
# inp neuron 1 fires once after that
indices = zeros(neurons_per_block + 1)
indices[-1] = 1
times = arange(neurons_per_block + 1) * default_dt
inp = SpikeGeneratorGroup(2, indices, times)
G = NeuronGroup(N, 'v:1', threshold='v>1', reset='v=0')
S = Synapses(inp, G, 'w:1', on_pre='v+=w')
S.connect()
# inp neuron 0 has no effect on target neurons (weight 0)
S.w['i == 0'] = 0
# inp neuron 1 makes G neurons spike
S.w['i == 1'] = 1.1
# delays from inp neuron 0 are equal to post neuron number j
S.delay[
'i == 0'] = '(j % neurons_per_block) * default_dt' # arange(2048) per block
# delays from inp neuron 0 are the same except of the delay for the synapse
# of the thread of the second loop in the spikequeue.push(), which is the
# same as the previous synapse
S.delay[
'i == 1'] = '(j % neurons_per_block) * default_dt' # arange(2048), but [..., 1023, 1023, 1025, ...]
S.delay[
'i == 1 and j % neurons_per_block == threads_per_block'] = '(threads_per_block - 1) * default_dt'
mon = SpikeMonitor(G)
# first neurons_per_block time steps we resize queues
# at time step neurons_per_block + 1 we send a spike that triggeres
# postsynaptic spikes with delays = arange(neurons_per_block), but [..., 1023, 1023, 1035, ...]
# then it takes neurons_per_block - 1 time steps for all effects to be applied
time_steps = 2 * neurons_per_block + 2
run(time_steps * default_dt)
assert_allclose(S.delay[0, :], delays0)
assert_allclose(S.delay[1, :], delays)
assert_equal(len(mon), N)
unique_mon_t = unique(mon.t)
for n in range(time_steps):
t = n * default_dt
if n <= neurons_per_block or n == time_steps - 1:
# no spikes
assert t not in unique_mon_t
else:
try:
mon_idx = sort(mon.i[mon.t == t])
# first neuron (idx 0) spikes at time step neurons_per_block + 1
spiking_neuron_idx = n - (neurons_per_block + 1)
# neuron indices increment over blocks, get idx starting at 0 per block
mon_idx_per_block = mod(mon_idx, neurons_per_block)
assert_array_equal(sort(mon_idx), mon_idx)
if spiking_neuron_idx == threads_per_block - 1: # 1023
# [1023, 1024, 1023, 1024, ...]
expected_indices = tile(
[spiking_neuron_idx, spiking_neuron_idx + 1],
num_blocks)
elif spiking_neuron_idx == threads_per_block: # 1024
# []
expected_indices = array([])
else:
# [spiking_neuron_idx, spiking_neuron_idx, ...]
expected_indices = tile([spiking_neuron_idx], num_blocks)
assert_equal(mon_idx_per_block, expected_indices)
except AssertionError:
print('n =', n)
print(mon_idx)
raise
def test_spatialneuron_morphology_assignment():
sec = Cylinder(length=50 * um, diameter=10 * um, n=2)
sec.sec1 = Cylinder(length=50 * um, diameter=10 * um, n=2)
sec.sec1.sec11 = Cylinder(length=50 * um, diameter=10 * um, n=2)
sec.sec1.sec12 = Cylinder(length=50 * um, diameter=10 * um, n=2)
sec.sec2 = Cylinder(length=50 * um, diameter=10 * um, n=2)
sec.sec2.sec21 = Cylinder(length=50 * um, diameter=10 * um, n=2)
neuron = SpatialNeuron(sec, 'Im = 0*amp/meter**2 : amp/meter**2')
neuron.v[sec.sec1.sec11] = 1 * volt
assert_allclose(neuron.sec1.sec11.v[:], np.ones(2) * volt)
assert_allclose(neuron.sec1.sec12.v[:], np.zeros(2) * volt)
assert_allclose(neuron.sec1.main.v[:], np.zeros(2) * volt)
assert_allclose(neuron.main.v[:], np.zeros(2) * volt)
assert_allclose(neuron.sec2.v[:], np.zeros(4) * volt)
neuron.v[sec.sec2[25 * um:]] = 2 * volt
neuron.v[sec.sec1[:25 * um]] = 3 * volt
assert_allclose(neuron.main.v[:], np.zeros(2) * volt)
assert_allclose(neuron.sec2.main.v[:], [0, 2] * volt)
assert_allclose(neuron.sec2.sec21.v[:], np.zeros(2) * volt)
assert_allclose(neuron.sec1.main.v[:], [3, 0] * volt)
assert_allclose(neuron.sec1.sec11.v[:], np.ones(2) * volt)
assert_allclose(neuron.sec1.sec12.v[:], np.zeros(2) * volt)
def test_construction_coordinates():
# Same as test_construction, but uses coordinates instead of lengths to
# set up everything
# Note that all coordinates here are relative to the origin of the
# respective cylinder
BrianLogger.suppress_name('resolution_conflict')
morpho = Soma(diameter=30 * um)
morpho.L = Cylinder(x=[0, 10] * um, diameter=1 * um, n=10)
morpho.LL = Cylinder(y=[0, 5] * um, diameter=2 * um, n=5)
morpho.LR = Cylinder(z=[0, 5] * um, diameter=2 * um, n=10)
morpho.right = Cylinder(x=[0, sqrt(2) * 1.5] * um,
y=[0, sqrt(2) * 1.5] * um,
diameter=1 * um,
n=7)
morpho.right.nextone = Cylinder(y=[0, sqrt(2)] * um,
z=[0, sqrt(2)] * um,
diameter=1 * um,
n=3)
gL = 1e-4 * siemens / cm**2
EL = -70 * mV
eqs = '''
Im=gL*(EL-v) : amp/meter**2
I : meter (point current)
'''
# Check units of currents
with pytest.raises(DimensionMismatchError):
SpatialNeuron(morphology=morpho, model=eqs)
eqs = '''
Im=gL*(EL-v) : amp/meter**2
'''
neuron = SpatialNeuron(morphology=morpho,
model=eqs,
Cm=1 * uF / cm**2,
Ri=100 * ohm * cm)
# Test initialization of values
neuron.LL.v = EL
assert_allclose(neuron.L.main.v, 0 * mV)
assert_allclose(neuron.LL.v, EL)
neuron.LL[1 * um:3 * um].v = 0 * mV
assert_allclose(neuron.LL.v, Quantity([EL, 0 * mV, 0 * mV, EL, EL]))
assert_allclose(neuron.Cm, 1 * uF / cm**2)
# Test morphological variables
assert_allclose(neuron.L.main.x, morpho.L.x)
assert_allclose(neuron.LL.main.x, morpho.LL.x)
assert_allclose(neuron.right.main.x, morpho.right.x)
assert_allclose(neuron.L.main.distance, morpho.L.distance)
# assert_allclose(neuron.L.main.diameter, morpho.L.diameter)
assert_allclose(neuron.L.main.area, morpho.L.area)
assert_allclose(neuron.L.main.length, morpho.L.length)
# Check basic consistency of the flattened representation
assert all(neuron.diffusion_state_updater._ends[:].flat >=
neuron.diffusion_state_updater._starts[:].flat)
# Check that length and distances make sense
assert_allclose(sum(morpho.L.length), 10 * um)
assert_allclose(morpho.L.distance, (0.5 + np.arange(10)) * um)
assert_allclose(sum(morpho.LL.length), 5 * um)
assert_allclose(morpho.LL.distance, (10 + .5 + np.arange(5)) * um)
assert_allclose(sum(morpho.LR.length), 5 * um)
assert_allclose(morpho.LR.distance, (10 + 0.25 + np.arange(10) * 0.5) * um)
assert_allclose(sum(morpho.right.length), 3 * um)
assert_allclose(morpho.right.distance, (0.5 + np.arange(7)) * 3. / 7. * um)
assert_allclose(sum(morpho.right.nextone.length), 2 * um)
assert_allclose(morpho.right.nextone.distance,
3 * um + (0.5 + np.arange(3)) * 2. / 3. * um)
def test_rall():
'''
Test simulation of a cylinder plus two branches, with diameters according to Rall's formula
'''
if prefs.core.default_float_dtype is np.float32:
pytest.skip('Need double precision for this test')
BrianLogger.suppress_name('resolution_conflict')
defaultclock.dt = 0.01 * ms
# Passive channels
gL = 1e-4 * siemens / cm**2
EL = -70 * mV
# Morphology
diameter = 1 * um
length = 300 * um
Cm = 1 * uF / cm**2
Ri = 150 * ohm * cm
N = 500
rm = 1 / (gL * pi * diameter) # membrane resistance per unit length
ra = (4 * Ri) / (pi * diameter**2) # axial resistance per unit length
la = sqrt(rm / ra) # space length
morpho = Cylinder(diameter=diameter, length=length, n=N)
d1 = 0.5 * um
L1 = 200 * um
rm = 1 / (gL * pi * d1) # membrane resistance per unit length
ra = (4 * Ri) / (pi * d1**2) # axial resistance per unit length
l1 = sqrt(rm / ra) # space length
morpho.L = Cylinder(diameter=d1, length=L1, n=N)
d2 = (diameter**1.5 - d1**1.5)**(1. / 1.5)
rm = 1 / (gL * pi * d2) # membrane resistance per unit length
ra = (4 * Ri) / (pi * d2**2) # axial resistance per unit length
l2 = sqrt(rm / ra) # space length
L2 = (L1 / l1) * l2
morpho.R = Cylinder(diameter=d2, length=L2, n=N)
eqs = '''
Im=gL*(EL-v) : amp/meter**2
I : amp (point current)
'''
neuron = SpatialNeuron(morphology=morpho, model=eqs, Cm=Cm, Ri=Ri)
neuron.v = EL
neuron.I[0] = 0.02 * nA # injecting at the left end
run(100 * ms)
# Check space constant calculation
assert_allclose(la, neuron.space_constant[0])
assert_allclose(l1, neuron.L.space_constant[0])
assert_allclose(l2, neuron.R.space_constant[0])
# Theory
x = neuron.main.distance
ra = la * 4 * Ri / (pi * diameter**2)
l = length / la + L1 / l1
theory = EL + ra * neuron.I[0] * cosh(l - x / la) / sinh(l)
v = neuron.main.v
assert_allclose(v - EL, theory - EL, rtol=1e12, atol=1e8)
x = neuron.L.distance
theory = EL + ra * neuron.I[0] * cosh(l - neuron.main.distance[-1] / la -
(x - neuron.main.distance[-1]) /
l1) / sinh(l)
v = neuron.L.v
assert_allclose(v - EL, theory - EL, rtol=1e12, atol=1e8)
x = neuron.R.distance
theory = EL + ra * neuron.I[0] * cosh(l - neuron.main.distance[-1] / la -
(x - neuron.main.distance[-1]) /
l2) / sinh(l)
v = neuron.R.v
assert_allclose(v - EL, theory - EL, rtol=1e12, atol=1e8)
def test_spatialneuron_subtree_assignment():
sec = Cylinder(length=50 * um, diameter=10 * um, n=2)
sec.sec1 = Cylinder(length=50 * um, diameter=10 * um, n=2)
sec.sec1.sec11 = Cylinder(length=50 * um, diameter=10 * um, n=2)
sec.sec1.sec12 = Cylinder(length=50 * um, diameter=10 * um, n=2)
sec.sec2 = Cylinder(length=50 * um, diameter=10 * um, n=2)
sec.sec2.sec21 = Cylinder(length=50 * um, diameter=10 * um, n=2)
neuron = SpatialNeuron(sec, 'Im = 0*amp/meter**2 : amp/meter**2')
neuron.v = 1 * volt
assert_allclose(neuron.v[:], np.ones(12) * volt)
neuron.sec1.v += 1 * volt
assert_allclose(neuron.main.v[:], np.ones(2) * volt)
assert_allclose(neuron.sec1.v[:], np.ones(6) * 2 * volt)
assert_allclose(neuron.sec1.main.v[:], np.ones(2) * 2 * volt)
assert_allclose(neuron.sec1.sec11.v[:], np.ones(2) * 2 * volt)
assert_allclose(neuron.sec1.sec12.v[:], np.ones(2) * 2 * volt)
assert_allclose(neuron.sec2.v[:], np.ones(4) * volt)
neuron.sec2.v = 5 * volt
assert_allclose(neuron.sec2.v[:], np.ones(4) * 5 * volt)
assert_allclose(neuron.sec2.main.v[:], np.ones(2) * 5 * volt)
assert_allclose(neuron.sec2.sec21.v[:], np.ones(2) * 5 * volt)
def test_rallpack2():
'''
Rallpack 2
'''
if prefs.core.default_float_dtype is np.float32:
pytest.skip('Need double precision for this test')
defaultclock.dt = 0.1 * ms
# Morphology
diameter = 32 * um
length = 16 * um
Cm = 1 * uF / cm**2
Ri = 100 * ohm * cm
# Construct binary tree according to Rall's formula
morpho = Cylinder(n=1, diameter=diameter, y=[0, float(length)] * meter)
endpoints = {morpho}
for depth in range(1, 10):
diameter /= 2.**(1. / 3.)
length /= 2.**(2. / 3.)
new_endpoints = set()
for endpoint in endpoints:
new_L = Cylinder(n=1, diameter=diameter, length=length)
new_R = Cylinder(n=1, diameter=diameter, length=length)
new_endpoints.add(new_L)
new_endpoints.add(new_R)
endpoint.L = new_L
endpoint.R = new_R
endpoints = new_endpoints
# Passive channels
gL = 1. / (40000 * ohm * cm**2)
EL = -65 * mV
eqs = '''
Im = gL*(EL - v) : amp/meter**2
I : amp (point current, constant)
'''
neuron = SpatialNeuron(morphology=morpho,
model=eqs,
Cm=Cm,
Ri=Ri,
method='rk4')
neuron.v = EL
neuron.I[0] = 0.1 * nA # injecting at the origin
endpoint_indices = [endpoint.indices[0] for endpoint in endpoints]
mon = StateMonitor(neuron,
'v',
record=[0] + endpoint_indices,
when='start',
dt=0.1 * ms)
run(250 * ms + defaultclock.dt)
# Load the theoretical results
basedir = os.path.join(os.path.abspath(os.path.dirname(__file__)),
'rallpack_data')
# Only use very second time step, since we run with 0.1ms instead of 0.05ms
data_0 = np.loadtxt(os.path.join(basedir, 'ref_branch.0'))[::2]
data_x = np.loadtxt(os.path.join(basedir, 'ref_branch.x'))[::2]
# sanity check: times are the same
assert_allclose(mon.t / second, data_0[:, 0])
assert_allclose(mon.t / second, data_x[:, 0])
# Check that all endpoints are the same:
for endpoint in endpoints:
assert_allclose(mon[endpoint].v, mon[endpoint[0]].v)
scale_0 = max(data_0[:, 1] * volt) - min(data_0[:, 1] * volt)
scale_x = max(data_x[:, 1] * volt) - min(data_x[:, 1] * volt)
squared_diff_0 = (data_0[:, 1] * volt - mon[0].v)**2
# One endpoint
squared_diff_x = (data_x[:, 1] * volt - mon[endpoint_indices[0]].v)**2
rel_RMS_0 = sqrt(mean(squared_diff_0)) / scale_0
rel_RMS_x = sqrt(mean(squared_diff_x)) / scale_x
max_rel_0 = sqrt(max(squared_diff_0)) / scale_0
max_rel_x = sqrt(max(squared_diff_x)) / scale_x
# RMS error should be < 0.25%, maximum error along the curve should be < 0.5%
assert 100 * rel_RMS_0 < 0.25
assert 100 * rel_RMS_x < 0.25
assert 100 * max_rel_0 < 0.5
assert 100 * max_rel_x < 0.5
def test_rallpack3():
'''
Rallpack 3
'''
if prefs.core.default_float_dtype is np.float32:
pytest.skip('Need double precision for this test')
defaultclock.dt = 1 * usecond
# Morphology
diameter = 1 * um
length = 1 * mm
N = 1000
morpho = Cylinder(diameter=diameter, length=length, n=N)
# Passive properties
gl = 1. / (40000 * ohm * cm**2)
El = -65 * mV
Cm = 1 * uF / cm**2
Ri = 100 * ohm * cm
# Active properties
ENa = 50 * mV
EK = -77 * mV
gNa = 120 * msiemens / cm**2
gK = 36 * msiemens / cm**2
eqs = '''
Im = gl * (El-v) + gNa * m**3 * h * (ENa-v) + gK * n**4 * (EK-v) : amp/meter**2
dm/dt = alpham * (1-m) - betam * m : 1
dn/dt = alphan * (1-n) - betan * n : 1
dh/dt = alphah * (1-h) - betah * h : 1
v_shifted = v - El : volt
alpham = (0.1/mV) * (-v_shifted+25*mV) / (exp((-v_shifted+25*mV) / (10*mV)) - 1)/ms : Hz
betam = 4 * exp(-v_shifted/(18*mV))/ms : Hz
alphah = 0.07 * exp(-v_shifted/(20*mV))/ms : Hz
betah = 1/(exp((-v_shifted+30*mV) / (10*mV)) + 1)/ms : Hz
alphan = (0.01/mV) * (-v_shifted+10*mV) / (exp((-v_shifted+10*mV) / (10*mV)) - 1)/ms : Hz
betan = 0.125*exp(-v_shifted/(80*mV))/ms : Hz
I : amp (point current, constant)
'''
axon = SpatialNeuron(morphology=morpho,
model=eqs,
Cm=Cm,
Ri=Ri,
method='exponential_euler')
axon.v = El
# Pre-calculated equilibrium values at v = El
axon.m = 0.0529324852572
axon.n = 0.317676914061
axon.h = 0.596120753508
axon.I[0] = 0.1 * nA # injecting at the left end
#Record at the two ends
mon = StateMonitor(axon, 'v', record=[0, 999], when='start', dt=0.05 * ms)
run(250 * ms + defaultclock.dt)
# Load the theoretical results
basedir = os.path.join(os.path.abspath(os.path.dirname(__file__)),
'rallpack_data')
data_0 = np.loadtxt(os.path.join(basedir, 'ref_axon.0.neuron'))
data_x = np.loadtxt(os.path.join(basedir, 'ref_axon.x.neuron'))
# sanity check: times are the same
assert_allclose(mon.t / second, data_0[:, 0])
assert_allclose(mon.t / second, data_x[:, 0])
scale_0 = max(data_0[:, 1] * volt) - min(data_0[:, 1] * volt)
scale_x = max(data_x[:, 1] * volt) - min(data_x[:, 1] * volt)
squared_diff_0 = (data_0[:, 1] * volt - mon[0].v)**2
squared_diff_x = (data_x[:, 1] * volt - mon[999].v)**2
rel_RMS_0 = sqrt(mean(squared_diff_0)) / scale_0
rel_RMS_x = sqrt(mean(squared_diff_x)) / scale_x
max_rel_0 = sqrt(max(squared_diff_0)) / scale_0
max_rel_x = sqrt(max(squared_diff_x)) / scale_x
# RMS error should be < 0.1%, maximum error along the curve should be < 0.5%
# Note that this is much stricter than the original Rallpack evaluation, but
# with the 1us time step, the voltage traces are extremely similar
assert 100 * rel_RMS_0 < 0.1
assert 100 * rel_RMS_x < 0.1
assert 100 * max_rel_0 < 0.5
assert 100 * max_rel_x < 0.5
def test_scale():
# Check that unit scaling is implemented correctly
from brian2.core.namespace import DEFAULT_UNITS
siprefixes = {
"y": 1e-24,
"z": 1e-21,
"a": 1e-18,
"f": 1e-15,
"p": 1e-12,
"n": 1e-9,
"u": 1e-6,
"m": 1e-3,
"": 1.0,
"k": 1e3,
"M": 1e6,
"G": 1e9,
"T": 1e12,
"P": 1e15,
"E": 1e18,
"Z": 1e21,
"Y": 1e24
}
for prefix in siprefixes:
if prefix in ['c', 'd', 'da', 'h']:
continue
scaled_unit = DEFAULT_UNITS[prefix + 'meter']
assert_allclose(float(scaled_unit), siprefixes[prefix])
assert_allclose(5 * scaled_unit / meter, 5 * siprefixes[prefix])
scaled_unit = DEFAULT_UNITS[prefix + 'meter2']
assert_allclose(float(scaled_unit), siprefixes[prefix]**2)
assert_allclose(5 * scaled_unit / meter2, 5 * siprefixes[prefix]**2)
scaled_unit = DEFAULT_UNITS[prefix + 'meter3']
assert_allclose(float(scaled_unit), siprefixes[prefix]**3)
assert_allclose(5 * scaled_unit / meter3, 5 * siprefixes[prefix]**3)
# liter, gram, and molar are special, they are not base units with a
# value of one, even though they do not have any prefix
for unit, factor in [('liter', 1e-3), ('litre', 1e-3), ('gram', 1e-3),
('gramme', 1e-3), ('molar', 1e3)]:
base_unit = DEFAULT_UNITS[unit]
scaled_unit = DEFAULT_UNITS[prefix + unit]
assert_allclose(float(scaled_unit), siprefixes[prefix] * factor)
assert_allclose(5 * scaled_unit / base_unit,
5 * siprefixes[prefix])
def test_array_cache():
# Check that variables are only accessible from Python when they should be
set_device('cpp_standalone', build_on_run=False)
G = NeuronGroup(10,
'''dv/dt = -v / (10*ms) : 1
w : 1
x : 1
y : 1
z : 1 (shared)''',
threshold='v>1')
S = Synapses(G, G, 'weight: 1', on_pre='w += weight')
S.connect(p=0.2)
S.weight = 7
# All neurongroup values should be known
assert_allclose(G.v, 0)
assert_allclose(G.w, 0)
assert_allclose(G.x, 0)
assert_allclose(G.y, 0)
assert_allclose(G.z, 0)
assert_allclose(G.i, np.arange(10))
# But the synaptic variable is not -- we don't know the number of synapses
with pytest.raises(NotImplementedError):
S.weight[:]
# Setting variables with explicit values should not change anything
G.v = np.arange(10) + 1
G.w = 2
G.y = 5
G.z = 7
assert_allclose(G.v, np.arange(10) + 1)
assert_allclose(G.w, 2)
assert_allclose(G.y, 5)
assert_allclose(G.z, 7)
# But setting with code should invalidate them
G.x = 'i*2'
with pytest.raises(NotImplementedError):
G.x[:]
# Make sure that the array cache does not allow to use incorrectly sized
# values to pass
with pytest.raises(ValueError):
setattr(G, 'w', [0, 2])
with pytest.raises(ValueError):
G.w.__setitem__(slice(0, 4), [0, 2])
run(10 * ms)
# v is now no longer known without running the network
with pytest.raises(NotImplementedError):
G.v[:]
# Neither is w, it is updated in the synapse
with pytest.raises(NotImplementedError):
G.w[:]
# However, no code touches y or z
assert_allclose(G.y, 5)
assert_allclose(G.z, 7)
# i is read-only anyway
assert_allclose(G.i, np.arange(10))
# After actually running the network, everything should be accessible
device.build(directory=None, with_output=False)
assert all(G.v > 0)
assert all(G.w > 0)
assert_allclose(G.x, np.arange(10) * 2)
assert_allclose(G.y, 5)
assert_allclose(G.z, 7)
assert_allclose(G.i, np.arange(10))
assert_allclose(S.weight, 7)
def test_math_functions():
'''
Test that math functions give the same result, regardless of whether used
directly or in generated Python or C++ code.
'''
default_dt = defaultclock.dt
test_array = np.array([-1, -0.5, 0, 0.5, 1])
def int_(x):
return array(x, dtype=int)
int_.__name__ = 'int'
with catch_logs() as _: # Let's suppress warnings about illegal values
# Functions with a single argument
for func in [
cos, tan, sinh, cosh, tanh, arcsin, arccos, arctan, log, log10,
exp, np.sqrt, np.ceil, np.floor, np.sign, int_
]:
# Calculate the result directly
numpy_result = func(test_array)
# Calculate the result in a somewhat complicated way by using a
# subexpression in a NeuronGroup
if func.__name__ == 'absolute':
# we want to use the name abs instead of absolute
func_name = 'abs'
else:
func_name = func.__name__
G = NeuronGroup(
len(test_array), '''func = {func}(variable) : 1
variable : 1'''.format(func=func_name))
G.variable = test_array
mon = StateMonitor(G, 'func', record=True)
net = Network(G, mon)
net.run(default_dt)
assert_allclose(
numpy_result,
mon.func_.flatten(),
err_msg='Function %s did not return the correct values' %
func.__name__)
# Functions/operators
scalar = 3
for func, operator in [(np.power, '**'), (np.mod, '%')]:
# Calculate the result directly
numpy_result = func(test_array, scalar)
# Calculate the result in a somewhat complicated way by using a
# subexpression in a NeuronGroup
G = NeuronGroup(
len(test_array), '''func = variable {op} scalar : 1
variable : 1'''.format(op=operator))
G.variable = test_array
mon = StateMonitor(G, 'func', record=True)
net = Network(G, mon)
net.run(default_dt)
assert_allclose(
numpy_result,
mon.func_.flatten(),
err_msg='Function %s did not return the correct values' %
func.__name__)
def test_rate_monitor_1():
G = NeuronGroup(5, 'v : 1', threshold='v>1') # no reset
G.v = 1.1 # All neurons spike every time step
rate_mon = PopulationRateMonitor(G)
run(10 * defaultclock.dt)
assert_allclose(rate_mon.t, np.arange(10) * defaultclock.dt)
assert_allclose(rate_mon.t_, np.arange(10) * defaultclock.dt_)
assert_allclose(rate_mon.t, np.arange(10) * defaultclock.dt)
assert_allclose(rate_mon.rate, np.ones(10) / defaultclock.dt)
assert_allclose(rate_mon.rate_, np.asarray(np.ones(10) / defaultclock.dt_))
# Check that indexing into the VariableView works (this fails if we do not
# update the N variable correctly)
assert_allclose(rate_mon.t[:5], np.arange(5) * defaultclock.dt)
def test_construction():
BrianLogger.suppress_name('resolution_conflict')
morpho = Soma(diameter=30*um)
morpho.L = Cylinder(length=10*um, diameter=1*um, n=10)
morpho.LL = Cylinder(length=5*um, diameter=2*um, n=5)
morpho.LR = Cylinder(length=5*um, diameter=2*um, n=10)
morpho.right = Cylinder(length=3*um, diameter=1*um, n=7)
morpho.right.nextone = Cylinder(length=2*um, diameter=1*um, n=3)
gL=1e-4*siemens/cm**2
EL=-70*mV
eqs="""
Im=gL*(EL-v) : amp/meter**2
I : meter (point current)
"""
# Check units of currents
with pytest.raises(DimensionMismatchError):
SpatialNeuron(morphology=morpho, model=eqs)
eqs="""
Im=gL*(EL-v) : amp/meter**2
"""
neuron = SpatialNeuron(morphology=morpho, model=eqs, Cm=1 * uF / cm ** 2, Ri=100 * ohm * cm)
# Test initialization of values
neuron.LL.v = EL
assert_allclose(neuron.L.main.v, 0*mV)
assert_allclose(neuron.LL.v, EL)
neuron.LL[1*um:3*um].v = 0*mV
assert_allclose(neuron.LL.v, Quantity([EL, 0*mV, 0*mV, EL, EL]))
assert_allclose(neuron.Cm, 1 * uF / cm ** 2)
# Test morphological variables
assert_allclose(neuron.L.main.distance, morpho.L.distance)
assert_allclose(neuron.L.main.area, morpho.L.area)
assert_allclose(neuron.L.main.length, morpho.L.length)
# Check basic consistency of the flattened representation
assert all(neuron.diffusion_state_updater._ends[:].flat >=
neuron.diffusion_state_updater._starts[:].flat)
# Check that length and distances make sense
assert_allclose(sum(morpho.L.length), 10*um)
assert_allclose(morpho.L.distance, (0.5 + np.arange(10))*um)
assert_allclose(sum(morpho.LL.length), 5*um)
assert_allclose(morpho.LL.distance, (10 + .5 + np.arange(5))*um)
assert_allclose(sum(morpho.LR.length), 5*um)
assert_allclose(morpho.LR.distance, (10 + 0.25 + np.arange(10)*0.5)*um)
assert_allclose(sum(morpho.right.length), 3*um)
assert_allclose(morpho.right.distance, (0.5 + np.arange(7))*3./7.*um)
assert_allclose(sum(morpho.right.nextone.length), 2*um)
assert_allclose(morpho.right.nextone.distance, 3*um + (0.5 + np.arange(3))*2./3.*um)
def test_atomics_parallelisation():
# Adapted from brian2.test_synapses:test_ufunc_at_vectorisation()
for n, code in enumerate(permutation_analysis_good_examples):
should_be_able_to_use_ufunc_at = not 'NOT_UFUNC_AT_VECTORISABLE' in code
if should_be_able_to_use_ufunc_at:
use_ufunc_at_list = [False, True]
else:
use_ufunc_at_list = [True]
code = deindent(code)
vars = get_identifiers(code)
vars_src = []
vars_tgt = []
vars_syn = []
vars_shared = []
vars_const = {}
for var in vars:
if var.endswith('_pre'):
vars_src.append(var[:-4])
elif var.endswith('_post'):
vars_tgt.append(var[:-5])
elif var.endswith('_syn'):
vars_syn.append(var[:-4])
elif var.endswith('_shared'):
vars_shared.append(var[:-7])
elif var.endswith('_const'):
vars_const[var[:-6]] = 42
eqs_src = '\n'.join(var + ':1' for var in vars_src)
eqs_tgt = '\n'.join(var + ':1' for var in vars_tgt)
eqs_syn = '\n'.join(var + ':1' for var in vars_syn)
eqs_syn += '\n' + '\n'.join(var + ':1 (shared)' for var in vars_shared)
origvals = {}
endvals = {}
group_size = 1000
syn_size = group_size**2
try:
BrianLogger._log_messages.clear()
with catch_logs(log_level=logging.INFO) as caught_logs:
for use_ufunc_at in use_ufunc_at_list:
set_device('cuda_standalone',
directory=None,
compile=True,
run=True,
debug=False)
CUDACodeGenerator._use_atomics = use_ufunc_at
src = NeuronGroup(group_size,
eqs_src,
threshold='True',
name='src')
tgt = NeuronGroup(group_size, eqs_tgt, name='tgt')
syn = Synapses(src,
tgt,
eqs_syn,
on_pre=code.replace('_syn', '').replace(
'_const', '').replace('_shared', ''),
name='syn',
namespace=vars_const)
syn.connect()
for G, vars in [(src, vars_src), (tgt, vars_tgt),
(syn, vars_syn)]:
for var in vars:
fullvar = var + G.name
if fullvar in origvals:
G.state(var)[:] = origvals[fullvar]
else:
if isinstance(G, Synapses):
val = rand(syn_size)
else:
val = rand(len(G))
G.state(var)[:] = val
origvals[fullvar] = val.copy()
Network(src, tgt, syn).run(5 * defaultclock.dt)
for G, vars in [(src, vars_src), (tgt, vars_tgt),
(syn, vars_syn)]:
for var in vars:
fullvar = var + G.name
val = G.state(var)[:].copy()
if fullvar in endvals:
assert_allclose(val,
endvals[fullvar],
err_msg='%d: %s' % (n, code),
rtol=1e-5)
else:
endvals[fullvar] = val
device.reinit()
device.activate()
cuda_generator_messages = [
l for l in caught_logs
if l[1] == 'brian2.codegen.generators.cuda_generator'
]
if should_be_able_to_use_ufunc_at:
assert len(
cuda_generator_messages) == 0, cuda_generator_messages
else:
assert len(
cuda_generator_messages) == 1, cuda_generator_messages
log_lev, log_mod, log_msg = cuda_generator_messages[0]
assert log_msg.startswith(
'Failed to parallelise code'), log_msg
finally:
CUDACodeGenerator._use_atomics = False #restore it
device.reinit()
device.activate()
def test_default_function_implementations():
''' Test that all default functions work as expected '''
# NeuronGroup variables are set in device code
# Synapses are generated in host code
# int function arguments use the template code
# double arguments use the overloaded function
### sin
# Promotes int to float in template device code
G = NeuronGroup(1, 'v: 1')
G.v = 'sin(i)'
# Uses int overloaded C++ function in template host code
S = Synapses(G, G, 'w: 1')
S.connect(condition='sin(i)==0')
# Uses double overloaded func (device code)
G1 = NeuronGroup(1, 'v: 1')
G1.v = 'sin(i*1.0)'
# Uses double overloaded func (host code)
S1 = Synapses(G, G, 'w: 1')
S1.connect(condition='sin(i*1.0)==0')
### cos
G2 = NeuronGroup(1, 'v: 1')
G2.v = 'cos(i)'
S2 = Synapses(G, G, 'w: 1')
S2.connect(condition='cos(i)==1')
G3 = NeuronGroup(1, 'v: 1')
G3.v = 'cos(i*1.0)'
S3 = Synapses(G, G, 'w: 1')
S3.connect(condition='cos(i*1.0)==1')
### tan
G4 = NeuronGroup(1, 'v: 1')
G4.v = 'tan(i)'
S4 = Synapses(G, G, 'w: 1')
S4.connect(condition='tan(i)==0')
G5 = NeuronGroup(1, 'v: 1')
G5.v = 'tan(i*1.0)'
S5 = Synapses(G, G, 'w: 1')
S5.connect(condition='tan(i*1.0)==0')
### sinh
G6 = NeuronGroup(1, 'v: 1')
G6.v = 'sinh(i)'
S6 = Synapses(G, G, 'w: 1')
S6.connect(condition='sinh(i)==0')
G7 = NeuronGroup(1, 'v: 1')
G7.v = 'sinh(i*1.0)'
S7 = Synapses(G, G, 'w: 1')
S7.connect(condition='sinh(i*1.0)==0')
### cosh
G8 = NeuronGroup(1, 'v: 1')
G8.v = 'cosh(i)'
S8 = Synapses(G, G, 'w: 1')
S8.connect(condition='cosh(i)==1')
G9 = NeuronGroup(1, 'v: 1')
G9.v = 'cosh(i*1.0)'
S9 = Synapses(G, G, 'w: 1')
S9.connect(condition='cosh(i*1.0)==1')
### tanh
G10 = NeuronGroup(1, 'v: 1')
G10.v = 'tanh(i)'
S10 = Synapses(G, G, 'w: 1')
S10.connect(condition='tanh(i)==0')
G11 = NeuronGroup(1, 'v: 1')
G11.v = 'tanh(i*1.0)'
S11 = Synapses(G, G, 'w: 1')
S11.connect(condition='tanh(i*1.0)==0')
### exp
G12 = NeuronGroup(1, 'v: 1')
G12.v = 'exp(i)'
S12 = Synapses(G, G, 'w: 1')
S12.connect(condition='exp(i)==1')
G13 = NeuronGroup(1, 'v: 1')
G13.v = 'exp(i*1.0)'
S13 = Synapses(G, G, 'w: 1')
S13.connect(condition='exp(i*1.0)==1')
### log
G14 = NeuronGroup(1, 'v: 1')
G14.v = 'log(i+1)'
S14 = Synapses(G, G, 'w: 1')
S14.connect(condition='log(i+1)==0')
G15 = NeuronGroup(1, 'v: 1')
G15.v = 'log(i+1.0)'
S15 = Synapses(G, G, 'w: 1')
S15.connect(condition='log(i+1.0)==0')
### log10
G16 = NeuronGroup(1, 'v: 1')
G16.v = 'log10(i+1)'
S16 = Synapses(G, G, 'w: 1')
S16.connect(condition='log10(i+1)==0')
G17 = NeuronGroup(1, 'v: 1')
G17.v = 'log10(i+1.0)'
S17 = Synapses(G, G, 'w: 1')
S17.connect(condition='log10(i+1.0)==0')
### sqrt
G18 = NeuronGroup(1, 'v: 1')
G18.v = 'sqrt(i)'
S18 = Synapses(G, G, 'w: 1')
S18.connect(condition='sqrt(i)==0')
G19 = NeuronGroup(1, 'v: 1')
G19.v = 'sqrt(i*1.0)'
S19 = Synapses(G, G, 'w: 1')
S19.connect(condition='sqrt(i*1.0)==0')
### ceil
G20 = NeuronGroup(1, 'v: 1')
G20.v = 'ceil(i)'
S20 = Synapses(G, G, 'w: 1')
S20.connect(condition='ceil(i)==0')
G21 = NeuronGroup(1, 'v: 1')
G21.v = 'ceil(i*1.0)'
S21 = Synapses(G, G, 'w: 1')
S21.connect(condition='ceil(i*1.0)==0')
### floor
G22 = NeuronGroup(1, 'v: 1')
G22.v = 'floor(i)'
S22 = Synapses(G, G, 'w: 1')
S22.connect(condition='floor(i)==0')
G23 = NeuronGroup(1, 'v: 1')
G23.v = 'floor(i*1.0)'
S23 = Synapses(G, G, 'w: 1')
S23.connect(condition='floor(i*1.0)==0')
### arcsin
G24 = NeuronGroup(1, 'v: 1')
G24.v = 'arcsin(i)'
S24 = Synapses(G, G, 'w: 1')
S24.connect(condition='arcsin(i)==0')
G25 = NeuronGroup(1, 'v: 1')
G25.v = 'arcsin(i*1.0)'
S25 = Synapses(G, G, 'w: 1')
S25.connect(condition='arcsin(i*1.0)==0')
### arccos
G26 = NeuronGroup(1, 'v: 1')
G26.v = 'arccos(i+1)'
S26 = Synapses(G, G, 'w: 1')
S26.connect(condition='arccos(i+1)==0')
G27 = NeuronGroup(1, 'v: 1')
G27.v = 'arccos(i+1.0)'
S27 = Synapses(G, G, 'w: 1')
S27.connect(condition='arccos(i+1.0)==0')
### arctan
G28 = NeuronGroup(1, 'v: 1')
G28.v = 'arctan(i)'
S28 = Synapses(G, G, 'w: 1')
S28.connect(condition='arctan(i)==0')
G29 = NeuronGroup(1, 'v: 1')
G29.v = 'arctan(i*1.0)'
S29 = Synapses(G, G, 'w: 1')
S29.connect(condition='arctan(i*1.0)==0')
### abs
G30 = NeuronGroup(1, 'v: 1')
G30.v = 'abs(i-1)'
S30 = Synapses(G, G, 'w: 1')
S30.connect(condition='abs(i-1)==1')
G31 = NeuronGroup(1, 'v: 1')
G31.v = 'abs(i-1.0)'
S31 = Synapses(G, G, 'w: 1')
S31.connect(condition='abs(i-1.0)==1')
### int
G32 = NeuronGroup(1, 'v: 1')
G32.v = 'int(i>0)'
S32 = Synapses(G, G, 'w: 1')
S32.connect(condition='int(i>0)==0')
G33 = NeuronGroup(1, 'v: 1')
G33.v = 'int(i+0.1)'
S33 = Synapses(G, G, 'w: 1')
S33.connect(condition='int(i+0.1)==0')
### clip
G34 = NeuronGroup(1, 'v: 1')
G34.v = 'clip(i+1,-1,0)'
S34 = Synapses(G, G, 'w: 1')
S34.connect(condition='clip(i+1,-1,0)==0')
G35 = NeuronGroup(1, 'v: 1')
G35.v = 'clip(i+1,-1.0,0.0)'
S35 = Synapses(G, G, 'w: 1')
S35.connect(condition='clip(i+1,-1.0,0.0)==0')
### sign
G36 = NeuronGroup(1, 'v: 1')
G36.v = 'sign(i+1)'
S36 = Synapses(G, G, 'w: 1')
S36.connect(condition='sign(i+1)==1')
### timestep
G38 = NeuronGroup(1, 'v: 1')
G38.v = 'timestep(0.1*ms, 0.001*ms)'
S38 = Synapses(G, G, 'w: 1')
S38.connect(condition='timestep(0.1*ms, 0.001*ms) == 100')
run(0 * ms)
assert_allclose([G38.v[0]], [100])
assert_allclose([S38.N[:]], [1])
assert_allclose([G36.v[0]], [1])
assert_allclose([S36.N[:]], [1])
assert_allclose([G34.v[0], G35.v[0]], [0, 0])
assert_allclose([S34.N[:], S35.N[:]], [1, 1])
assert_allclose([G32.v[0], G33.v[0]], [0, 0])
assert_allclose([S32.N[:], S33.N[:]], [1, 1])
assert_allclose([G30.v[0], G31.v[0]], [1, 1])
assert_allclose([S30.N[:], S31.N[:]], [1, 1])
assert_allclose([G28.v[0], G29.v[0]], [0, 0])
assert_allclose([S28.N[:], S29.N[:]], [1, 1])
assert_allclose([G26.v[0], G27.v[0]], [0, 0])
assert_allclose([S26.N[:], S27.N[:]], [1, 1])
assert_allclose([G24.v[0], G25.v[0]], [0, 0])
assert_allclose([S24.N[:], S25.N[:]], [1, 1])
assert_allclose([G22.v[0], G23.v[0]], [0, 0])
assert_allclose([S22.N[:], S23.N[:]], [1, 1])
assert_allclose([G20.v[0], G21.v[0]], [0, 0])
assert_allclose([S20.N[:], S21.N[:]], [1, 1])
assert_allclose([G18.v[0], G19.v[0]], [0, 0])
assert_allclose([S18.N[:], S19.N[:]], [1, 1])
assert_allclose([G16.v[0], G17.v[0]], [0, 0])
assert_allclose([S16.N[:], S17.N[:]], [1, 1])
assert_allclose([G14.v[0], G15.v[0]], [0, 0])
assert_allclose([S14.N[:], S15.N[:]], [1, 1])
assert_allclose([G12.v[0], G13.v[0]], [1, 1])
assert_allclose([S12.N[:], S13.N[:]], [1, 1])
assert_allclose([G10.v[0], G11.v[0]], [0, 0])
assert_allclose([S10.N[:], S11.N[:]], [1, 1])
assert_allclose([G8.v[0], G9.v[0]], [1, 1])
assert_allclose([S8.N[:], S9.N[:]], [1, 1])
assert_allclose([G6.v[0], G7.v[0]], [0, 0])
assert_allclose([S6.N[:], S7.N[:]], [1, 1])
assert_allclose([G4.v[0], G5.v[0]], [0, 0])
assert_allclose([S4.N[:], S5.N[:]], [1, 1])
assert_allclose([G2.v[0], G3.v[0]], [1, 1])
assert_allclose([S2.N[:], S3.N[:]], [1, 1])
assert_allclose([G.v[0], G1.v[0]], [0, 0])
assert_allclose([S.N[:], S1.N[:]], [1, 1])
def test_binomial_values():
if prefs.core.default_float_dtype is np.float32:
# TODO: Make test single-precision compatible, see #262
pytest.skip('Need double precision for this test')
# On Denis' local computer this test blows up all available RAM + SWAP when
# compiling with all threads in parallel. Use half the threads instead.
import socket
if socket.gethostname() == 'selene':
prefs.devices.cpp_standalone.extra_make_args_unix = ['-j4']
my_f_approximated = BinomialFunction(100, 0.1, approximate=True)
my_f = BinomialFunction(100, 0.1, approximate=False)
# Test neurongroup every tick objects
G = NeuronGroup(10,
'''dx/dt = my_f_approximated()/ms: 1
dy/dt = my_f()/ms: 1''',
threshold='True')
G.run_regularly('''x = my_f_approximated()
y = my_f()''')
# Test synapses every tick objects (where N is not known at compile time)
syn = Synapses(
G,
G,
model='''
dw/dt = my_f()/ms: 1
dv/dt = my_f_approximated()/ms: 1
''',
on_pre='''
x += w
y += v
'''
# TODO: fails when having binomial here, why?
#on_pre='''
# x += w * my_f()
# y += v * my_f_approximated()
# '''
)
# Test synapses generation, which needs host side binomial
syn.connect(condition='my_f_approximated() < 100')
mon = StateMonitor(G, ['x', 'y'], record=True)
w_mon = StateMonitor(syn, ['w', 'v'], record=np.arange(100))
def init_group_variables(my_f, my_f_approximated):
# Test codeobjects run only once outside the network,
# G.x uses group_variable_set_conditional, G.x[:5] uses group_variable_set
# Synapse objects (N not known at compile time)
syn.w = 'my_f()'
syn.w['i < j'] = 'my_f_approximated()'
syn.v = 'my_f_approximated()'
syn.v['i < j'] = 'my_f()'
# Neurongroup object
G.x = 'my_f_approximated()'
G.x[:5] = 'my_f()'
G.y = 'my_f()'
G.y[:5] = 'my_f_approximated()'
# first run
init_group_variables(my_f, my_f_approximated)
run(2 * defaultclock.dt)
# second run
seed(11400)
init_group_variables(my_f, my_f_approximated)
run(2 * defaultclock.dt)
# third run
seed()
init_group_variables(my_f, my_f_approximated)
run(2 * defaultclock.dt)
# forth run
seed(11400)
init_group_variables(my_f, my_f_approximated)
run(2 * defaultclock.dt)
device.build(direct_call=False, **device.build_options)
run_values_1 = np.vstack([
mon.x[:, [0, 1]], mon.y[:, [0, 1]], w_mon.w[:, [0, 1]], w_mon.v[:,
[0, 1]]
])
run_values_2 = np.vstack([
mon.x[:, [2, 3]], mon.y[:, [2, 3]], w_mon.w[:, [2, 3]], w_mon.v[:,
[2, 3]]
])
run_values_3 = np.vstack([
mon.x[:, [4, 5]], mon.y[:, [4, 5]], w_mon.w[:, [4, 5]], w_mon.v[:,
[4, 5]]
])
run_values_4 = np.vstack([
mon.x[:, [6, 7]], mon.y[:, [6, 7]], w_mon.w[:, [6, 7]], w_mon.v[:,
[6, 7]]
])
# Two calls to binomial functions should return different numbers in all runs
for n, values in enumerate(
[run_values_1, run_values_2, run_values_3, run_values_4]):
with pytest.raises(AssertionError):
assert_allclose(values[:, 0], values[:, 1])
# 2. and 4. run set the same seed
assert_allclose(run_values_2, run_values_4)
# all other combinations should be different
with pytest.raises(AssertionError):
assert_allclose(run_values_1, run_values_2)
with pytest.raises(AssertionError):
assert_allclose(run_values_1, run_values_3)
with pytest.raises(AssertionError):
assert_allclose(run_values_2, run_values_3)
def test_rallpack1():
'''
Rallpack 1
'''
if prefs.core.default_float_dtype is np.float32:
pytest.skip('Need double precision for this test')
defaultclock.dt = 0.05 * ms
# Morphology
diameter = 1 * um
length = 1 * mm
Cm = 1 * uF / cm**2
Ri = 100 * ohm * cm
N = 1000
morpho = Cylinder(diameter=diameter, length=length, n=N)
# Passive channels
gL = 1. / (40000 * ohm * cm**2)
EL = -65 * mV
eqs = '''
Im = gL*(EL - v) : amp/meter**2
I : amp (point current, constant)
'''
neuron = SpatialNeuron(morphology=morpho, model=eqs, Cm=Cm, Ri=Ri)
neuron.v = EL
neuron.I[0] = 0.1 * nA # injecting at the left end
#Record at the two ends
mon = StateMonitor(neuron,
'v',
record=[0, 999],
when='start',
dt=0.05 * ms)
run(250 * ms + defaultclock.dt)
# Load the theoretical results
basedir = os.path.join(os.path.abspath(os.path.dirname(__file__)),
'rallpack_data')
data_0 = np.loadtxt(os.path.join(basedir, 'ref_cable.0'))
data_x = np.loadtxt(os.path.join(basedir, 'ref_cable.x'))
scale_0 = max(data_0[:, 1] * volt) - min(data_0[:, 1] * volt)
scale_x = max(data_x[:, 1] * volt) - min(data_x[:, 1] * volt)
squared_diff_0 = (data_0[:, 1] * volt - mon[0].v)**2
squared_diff_x = (data_x[:, 1] * volt - mon[999].v)**2
rel_RMS_0 = sqrt(mean(squared_diff_0)) / scale_0
rel_RMS_x = sqrt(mean(squared_diff_x)) / scale_x
max_rel_0 = sqrt(max(squared_diff_0)) / scale_0
max_rel_x = sqrt(max(squared_diff_x)) / scale_x
# sanity check: times are the same
assert_allclose(mon.t / second, data_0[:, 0])
assert_allclose(mon.t / second, data_x[:, 0])
# RMS error should be < 0.1%, maximum error along the curve should be < 0.5%
assert 100 * rel_RMS_0 < 0.1
assert 100 * rel_RMS_x < 0.1
assert 100 * max_rel_0 < 0.5
assert 100 * max_rel_x < 0.5
def test_spike_monitor():
G = NeuronGroup(3,
'''dv/dt = rate : 1
rate: Hz''',
threshold='v>1',
reset='v=0')
# We don't use 100 and 1000Hz, because then the membrane potential would
# be exactly at 1 after 10 resp. 100 timesteps. Due to floating point
# issues this will not be exact,
G.rate = [101, 0, 1001] * Hz
mon = SpikeMonitor(G)
with pytest.raises(ValueError):
SpikeMonitor(G, order=1) # need to specify 'when' as well
# Creating a SpikeMonitor for a Synapses object should not work
S = Synapses(G, G, on_pre='v += 0')
S.connect()
with pytest.raises(TypeError):
SpikeMonitor(S)
run(10 * ms)
spike_trains = mon.spike_trains()
assert_allclose(mon.t[mon.i == 0], [9.9] * ms)
assert len(mon.t[mon.i == 1]) == 0
assert_allclose(mon.t[mon.i == 2], np.arange(10) * ms + 0.9 * ms)
assert_allclose(mon.t_[mon.i == 0], np.array([9.9 * float(ms)]))
assert len(mon.t_[mon.i == 1]) == 0
assert_allclose(mon.t_[mon.i == 2], (np.arange(10) + 0.9) * float(ms))
assert_allclose(spike_trains[0], [9.9] * ms)
assert len(spike_trains[1]) == 0
assert_allclose(spike_trains[2], np.arange(10) * ms + 0.9 * ms)
assert_array_equal(mon.count, np.array([1, 0, 10]))
i, t = mon.it
i_, t_ = mon.it_
assert_array_equal(i, mon.i)
assert_array_equal(i, i_)
assert_array_equal(t, mon.t)
assert_array_equal(t_, mon.t_)
with pytest.raises(KeyError):
spike_trains[3]
with pytest.raises(KeyError):
spike_trains[-1]
with pytest.raises(KeyError):
spike_trains['string']
# Check that indexing into the VariableView works (this fails if we do not
# update the N variable correctly)
assert_allclose(mon.t[:5], [0.9, 1.9, 2.9, 3.9, 4.9] * ms)
def test_openmp_consistency():
previous_device = get_device()
n_cells = 100
n_recorded = 10
numpy.random.seed(42)
taum = 20 * ms
taus = 5 * ms
Vt = -50 * mV
Vr = -60 * mV
El = -49 * mV
fac = (60 * 0.27 / 10)
gmax = 20 * fac
dApre = .01
taupre = 20 * ms
taupost = taupre
dApost = -dApre * taupre / taupost * 1.05
dApost *= 0.1 * gmax
dApre *= 0.1 * gmax
connectivity = numpy.random.randn(n_cells, n_cells)
sources = numpy.random.randint(0, n_cells - 1, 10 * n_cells)
# Only use one spike per time step (to rule out that a single source neuron
# has more than one spike in a time step)
times = numpy.random.choice(
numpy.arange(10 * n_cells), 10 * n_cells, replace=False) * ms
v_init = Vr + numpy.random.rand(n_cells) * (Vt - Vr)
eqs = Equations('''
dv/dt = (g-(v-El))/taum : volt
dg/dt = -g/taus : volt
''')
results = {}
for (n_threads,
devicename) in [(0, 'runtime'), (0, 'cpp_standalone'),
(1, 'cpp_standalone'), (2, 'cpp_standalone'),
(3, 'cpp_standalone'), (4, 'cpp_standalone')]:
set_device(devicename, build_on_run=False, with_output=False)
Synapses.__instances__().clear()
if devicename == 'cpp_standalone':
reinit_and_delete()
prefs.devices.cpp_standalone.openmp_threads = n_threads
P = NeuronGroup(n_cells,
model=eqs,
threshold='v>Vt',
reset='v=Vr',
refractory=5 * ms)
Q = SpikeGeneratorGroup(n_cells, sources, times)
P.v = v_init
P.g = 0 * mV
S = Synapses(P,
P,
model='''dApre/dt=-Apre/taupre : 1 (event-driven)
dApost/dt=-Apost/taupost : 1 (event-driven)
w : 1''',
pre='''g += w*mV
Apre += dApre
w = w + Apost''',
post='''Apost += dApost
w = w + Apre''')
S.connect()
S.w = fac * connectivity.flatten()
T = Synapses(Q, P, model="w : 1", on_pre="g += w*mV")
T.connect(j='i')
T.w = 10 * fac
spike_mon = SpikeMonitor(P)
rate_mon = PopulationRateMonitor(P)
state_mon = StateMonitor(S,
'w',
record=np.arange(n_recorded),
dt=0.1 * second)
v_mon = StateMonitor(P, 'v', record=np.arange(n_recorded))
run(0.2 * second, report='text')
if devicename == 'cpp_standalone':
device.build(directory=None, with_output=False)
results[n_threads, devicename] = {}
results[n_threads, devicename]['w'] = state_mon.w
results[n_threads, devicename]['v'] = v_mon.v
results[n_threads, devicename]['s'] = spike_mon.num_spikes
results[n_threads, devicename]['r'] = rate_mon.rate[:]
for key1, key2 in [((0, 'runtime'), (0, 'cpp_standalone')),
((1, 'cpp_standalone'), (0, 'cpp_standalone')),
((2, 'cpp_standalone'), (0, 'cpp_standalone')),
((3, 'cpp_standalone'), (0, 'cpp_standalone')),
((4, 'cpp_standalone'), (0, 'cpp_standalone'))]:
assert_allclose(results[key1]['w'], results[key2]['w'])
assert_allclose(results[key1]['v'], results[key2]['v'])
assert_allclose(results[key1]['r'], results[key2]['r'])
assert_allclose(results[key1]['s'], results[key2]['s'])
reset_device(previous_device)
def test_state_variables():
'''
Test the setting and accessing of state variables in subgroups.
'''
G = NeuronGroup(10, 'v : volt')
SG = G[4:9]
with pytest.raises(DimensionMismatchError):
SG.__setattr__('v', -70)
SG.v_ = float(-80 * mV)
assert_allclose(G.v,
np.array([0, 0, 0, 0, -80, -80, -80, -80, -80, 0]) * mV)
assert_allclose(SG.v, np.array([-80, -80, -80, -80, -80]) * mV)
assert_allclose(
G.v_,
np.array([0, 0, 0, 0, -80, -80, -80, -80, -80, 0]) * float(mV))
assert_allclose(SG.v_, np.array([-80, -80, -80, -80, -80]) * float(mV))
# You should also be able to set variables with a string
SG.v = 'v + i*mV'
assert_allclose(SG.v[0], -80 * mV)
assert_allclose(SG.v[4], -76 * mV)
assert_allclose(G.v[4:9], -80 * mV + np.arange(5) * mV)
# Calculating with state variables should work too
assert all(G.v[4:9] - SG.v == 0)
# And in-place modification should work as well
SG.v += 10 * mV
assert_allclose(G.v[4:9], -70 * mV + np.arange(5) * mV)
SG.v *= 2
assert_allclose(G.v[4:9], 2 * (-70 * mV + np.arange(5) * mV))
# with unit checking
with pytest.raises(DimensionMismatchError):
SG.v.__iadd__(3 * second)
with pytest.raises(DimensionMismatchError):
SG.v.__iadd__(3)
with pytest.raises(DimensionMismatchError):
SG.v.__imul__(3 * second)
# Indexing with subgroups
assert_equal(G.v[SG], SG.v[:])
def test_spike_monitor_subgroups():
G = NeuronGroup(6, '''do_spike : boolean''', threshold='do_spike')
G.do_spike = [True, False, False, False, True, True]
spikes_all = SpikeMonitor(G)
spikes_1 = SpikeMonitor(G[:2])
spikes_2 = SpikeMonitor(G[2:4])
spikes_3 = SpikeMonitor(G[4:])
run(defaultclock.dt)
assert_allclose(spikes_all.i, [0, 4, 5])
assert_allclose(spikes_all.t, [0, 0, 0] * ms)
assert_allclose(spikes_1.i, [0])
assert_allclose(spikes_1.t, [0] * ms)
assert len(spikes_2.i) == 0
assert len(spikes_2.t) == 0
assert_allclose(spikes_3.i, [0, 1]) # recorded spike indices are relative
assert_allclose(spikes_3.t, [0, 0] * ms)
def test_time_after_run():
set_device('cpp_standalone', build_on_run=False)
# Check that the clock and network time after a run is correct, even if we
# have not actually run the code yet (via build)
G = NeuronGroup(10, 'dv/dt = -v/(10*ms) : 1')
net = Network(G)
assert_allclose(defaultclock.dt, 0.1 * ms)
assert_allclose(defaultclock.t, 0. * ms)
assert_allclose(G.t, 0. * ms)
assert_allclose(net.t, 0. * ms)
net.run(10 * ms)
assert_allclose(defaultclock.t, 10. * ms)
assert_allclose(G.t, 10. * ms)
assert_allclose(net.t, 10. * ms)
net.run(10 * ms)
assert_allclose(defaultclock.t, 20. * ms)
assert_allclose(G.t, 20. * ms)
assert_allclose(net.t, 20. * ms)
device.build(directory=None, with_output=False)
# Everything should of course still be accessible
assert_allclose(defaultclock.t, 20. * ms)
assert_allclose(G.t, 20. * ms)
assert_allclose(net.t, 20. * ms)
reset_device()
def test_event_monitor():
G = NeuronGroup(3,
'''dv/dt = rate : 1
rate: Hz''',
events={'my_event': 'v>1'})
G.run_on_event('my_event', 'v=0')
# We don't use 100 and 1000Hz, because then the membrane potential would
# be exactly at 1 after 10 resp. 100 timesteps. Due to floating point
# issues this will not be exact,
G.rate = [101, 0, 1001] * Hz
mon = EventMonitor(G, 'my_event')
net = Network(G, mon)
net.run(10 * ms)
event_trains = mon.event_trains()
assert_allclose(mon.t[mon.i == 0], [9.9] * ms)
assert len(mon.t[mon.i == 1]) == 0
assert_allclose(mon.t[mon.i == 2], np.arange(10) * ms + 0.9 * ms)
assert_allclose(mon.t_[mon.i == 0], np.array([9.9 * float(ms)]))
assert len(mon.t_[mon.i == 1]) == 0
assert_allclose(mon.t_[mon.i == 2], (np.arange(10) + 0.9) * float(ms))
assert_allclose(event_trains[0], [9.9] * ms)
assert len(event_trains[1]) == 0
assert_allclose(event_trains[2], np.arange(10) * ms + 0.9 * ms)
assert_array_equal(mon.count, np.array([1, 0, 10]))
i, t = mon.it
i_, t_ = mon.it_
assert_array_equal(i, mon.i)
assert_array_equal(i, i_)
assert_array_equal(t, mon.t)
assert_array_equal(t_, mon.t_)
with pytest.raises(KeyError):
event_trains[3]
with pytest.raises(KeyError):
event_trains[-1]
with pytest.raises(KeyError):
event_trains['string']
def test_storing_loading():
set_device('cpp_standalone', build_on_run=False)
G = NeuronGroup(
10, '''v : volt
x : 1
n : integer
b : boolean''')
v = np.arange(10) * volt
x = np.arange(10, 20)
n = np.arange(20, 30)
b = np.array([True, False]).repeat(5)
G.v = v
G.x = x
G.n = n
G.b = b
S = Synapses(
G, G, '''v_syn : volt
x_syn : 1
n_syn : integer
b_syn : boolean''')
S.connect(j='i')
S.v_syn = v
S.x_syn = x
S.n_syn = n
S.b_syn = b
run(0 * ms)
device.build(directory=None, with_output=False)
assert_allclose(G.v[:], v)
assert_allclose(S.v_syn[:], v)
assert_allclose(G.x[:], x)
assert_allclose(S.x_syn[:], x)
assert_allclose(G.n[:], n)
assert_allclose(S.n_syn[:], n)
assert_allclose(G.b[:], b)
assert_allclose(S.b_syn[:], b)
reset_device()
def test_state_monitor():
# Unique name to get the warning even for repeated runs of the test
unique_name = 'neurongroup_' + str(uuid.uuid4()).replace('-', '_')
# Check that all kinds of variables can be recorded
G = NeuronGroup(2,
'''dv/dt = -v / (10*ms) : 1
f = clip(v, 0.1, 0.9) : 1
rate: Hz''',
threshold='v>1',
reset='v=0',
refractory=2 * ms,
name=unique_name)
G.rate = [100, 1000] * Hz
G.v = 1
S = Synapses(G, G, 'w: siemens')
S.connect(True)
S.w = 'j*nS'
# A bit peculiar, but in principle one should be allowed to record
# nothing except for the time
nothing_mon = StateMonitor(G, [], record=True)
no_record = StateMonitor(G, 'v', record=False)
# Use a single StateMonitor
v_mon = StateMonitor(G, 'v', record=True)
v_mon1 = StateMonitor(G, 'v', record=[1])
# Use a StateMonitor for specified variables
multi_mon = StateMonitor(G, ['v', 'f', 'rate'], record=True)
multi_mon1 = StateMonitor(G, ['v', 'f', 'rate'], record=[1])
# Use a StateMonitor recording everything
all_mon = StateMonitor(G, True, record=True)
# Record synapses with explicit indices (the only way allowed in standalone)
synapse_mon = StateMonitor(S, 'w', record=np.arange(len(G)**2))
run(10 * ms)
# Check time recordings
assert_array_equal(nothing_mon.t, v_mon.t)
assert_array_equal(nothing_mon.t_, np.asarray(nothing_mon.t))
assert_array_equal(nothing_mon.t_, v_mon.t_)
assert_allclose(nothing_mon.t,
np.arange(len(nothing_mon.t)) * defaultclock.dt)
assert_array_equal(no_record.t, v_mon.t)
# Check v recording
assert_allclose(v_mon.v.T, np.exp(np.tile(-v_mon.t, (2, 1)).T / (10 * ms)))
assert_allclose(v_mon.v_.T,
np.exp(np.tile(-v_mon.t_, (2, 1)).T / float(10 * ms)))
assert_array_equal(v_mon.v, multi_mon.v)
assert_array_equal(v_mon.v_, multi_mon.v_)
assert_array_equal(v_mon.v, all_mon.v)
assert_array_equal(v_mon.v[1:2], v_mon1.v)
assert_array_equal(multi_mon.v[1:2], multi_mon1.v)
assert len(no_record.v) == 0
# Other variables
assert_array_equal(
multi_mon.rate_.T,
np.tile(np.atleast_2d(G.rate_), (multi_mon.rate.shape[1], 1)))
assert_array_equal(multi_mon.rate[1:2], multi_mon1.rate)
assert_allclose(np.clip(multi_mon.v, 0.1, 0.9), multi_mon.f)
assert_allclose(np.clip(multi_mon1.v, 0.1, 0.9), multi_mon1.f)
assert all(all_mon[0].not_refractory[:] == True), (
'not_refractory should '
'be True, but got'
'(not_refractory, v):'
'%s ' % str((all_mon.not_refractory, all_mon.v)))
# Synapses
assert_allclose(synapse_mon.w[:],
np.tile(S.j[:] * nS, (synapse_mon.w[:].shape[1], 1)).T)