def test_loop():
'''
Somewhat realistic test with a loop of magic networks
'''
def run_simulation():
G = NeuronGroup(10, 'dv/dt = -v / (10*ms) : 1',
reset='v=0', threshold='v>1')
G.v = np.linspace(0, 1, 10)
run(1*ms)
# We return potentially problematic references to a VariableView
return G.v
# First run
with catch_logs(log_level=logging.INFO) as l:
v = run_simulation()
assert v[0] == 0 and 0 < v[-1] < 1
# Check the debug messages for the number of included objects
magic_objects = [msg[2] for msg in l
if msg[1] == 'brian2.core.magic.magic_objects'][0]
assert '4 objects' in magic_objects
# Second run
with catch_logs(log_level=logging.INFO) as l:
v = run_simulation()
assert v[0] == 0 and 0 < v[-1] < 1
# Check the debug messages for the number of included objects
magic_objects = [msg[2] for msg in l
if msg[1] == 'brian2.core.magic.magic_objects'][0]
assert '4 objects' in magic_objects
def test_locally_constant_check():
default_dt = defaultclock.dt
# The linear state update can handle additive time-dependent functions
# (e.g. a TimedArray) but only if it can be safely assumed that the function
# is constant over a single time check
ta0 = TimedArray(np.array([1]), dt=default_dt) # ok
ta1 = TimedArray(np.array([1]), dt=2*default_dt) # ok
ta2 = TimedArray(np.array([1]), dt=default_dt/2) # not ok
ta3 = TimedArray(np.array([1]), dt=default_dt*1.5) # not ok
for ta_func, ok in zip([ta0, ta1, ta2, ta3], [True, True, False, False]):
# additive
G = NeuronGroup(1, 'dv/dt = -v/(10*ms) + ta(t)*Hz : 1',
method='linear', namespace={'ta': ta_func})
net = Network(G)
if ok:
# This should work
net.run(0*ms)
else:
# This should not
with catch_logs():
assert_raises(UnsupportedEquationsException, lambda: net.run(0*ms))
# multiplicative
G = NeuronGroup(1, 'dv/dt = -v*ta(t)/(10*ms) : 1',
method='linear', namespace={'ta': ta_func})
net = Network(G)
if ok:
# This should work
net.run(0*ms)
else:
# This should not
with catch_logs():
assert_raises(UnsupportedEquationsException, lambda: net.run(0*ms))
# If the argument is more than just "t", we cannot guarantee that it is
# actually locally constant
G = NeuronGroup(1, 'dv/dt = -v*ta(t/2.0)/(10*ms) : 1',
method='linear', namespace={'ta': ta0})
net = Network(G)
assert_raises(UnsupportedEquationsException, lambda: net.run(0*ms))
# Arbitrary functions are not constant over a time step
G = NeuronGroup(1, 'dv/dt = -v/(10*ms) + sin(2*pi*100*Hz*t)*Hz : 1',
method='linear')
net = Network(G)
assert_raises(UnsupportedEquationsException, lambda: net.run(0*ms))
# Neither is "t" itself
G = NeuronGroup(1, 'dv/dt = -v/(10*ms) + t/second**2 : 1', method='linear')
net = Network(G)
assert_raises(UnsupportedEquationsException, lambda: net.run(0*ms))
# But if the argument is not referring to t, all should be well
G = NeuronGroup(1, 'dv/dt = -v/(10*ms) + sin(2*pi*100*Hz*5*second)*Hz : 1',
method='linear')
net = Network(G)
net.run(0*ms)
def test_locally_constant_check():
default_dt = defaultclock.dt
# The linear state update can handle additive time-dependent functions
# (e.g. a TimedArray) but only if it can be safely assumed that the function
# is constant over a single time check
ta0 = TimedArray(np.array([1]), dt=default_dt) # ok
ta1 = TimedArray(np.array([1]), dt=2*default_dt) # ok
ta2 = TimedArray(np.array([1]), dt=default_dt/2) # not ok
ta3 = TimedArray(np.array([1]), dt=default_dt*1.5) # not ok
for ta_func, ok in zip([ta0, ta1, ta2, ta3], [True, True, False, False]):
# additive
G = NeuronGroup(1, 'dv/dt = -v/(10*ms) + ta(t)*Hz : 1',
method='linear', namespace={'ta': ta_func})
net = Network(G)
if ok:
# This should work
net.run(0*ms)
else:
# This should not
with catch_logs():
assert_raises(ValueError, lambda: net.run(0*ms))
# multiplicative
G = NeuronGroup(1, 'dv/dt = -v*ta(t)/(10*ms) : 1',
method='linear', namespace={'ta': ta_func})
net = Network(G)
if ok:
# This should work
net.run(0*ms)
else:
# This should not
with catch_logs():
assert_raises(ValueError, lambda: net.run(0*ms))
# If the argument is more than just "t", we cannot guarantee that it is
# actually locally constant
G = NeuronGroup(1, 'dv/dt = -v*ta(t/2.0)/(10*ms) : 1',
method='linear', namespace={'ta': ta0})
net = Network(G)
assert_raises(ValueError, lambda: net.run(0*ms))
# Arbitrary functions are not constant over a time step
G = NeuronGroup(1, 'dv/dt = -v/(10*ms) + sin(2*pi*100*Hz*t)*Hz : 1',
method='linear')
net = Network(G)
assert_raises(ValueError, lambda: net.run(0*ms))
# Neither is "t" itself
G = NeuronGroup(1, 'dv/dt = -v/(10*ms) + t/second**2 : 1', method='linear')
net = Network(G)
assert_raises(ValueError, lambda: net.run(0*ms))
# But if the argument is not referring to t, all should be well
G = NeuronGroup(1, 'dv/dt = -v/(10*ms) + sin(2*pi*100*Hz*5*second)*Hz : 1',
method='linear')
net = Network(G)
net.run(0*ms)
def test_set_interval_warning():
clock = Clock(dt=0.1*ms)
with catch_logs() as logs:
clock.set_interval(0*second, 1000*second) # no problem
assert len(logs) == 0
with catch_logs() as logs:
clock.set_interval(0*second, 10000000000*second) # too long
assert len(logs) == 1
assert logs[0][1].endswith('many_timesteps')
def test_schedule_warning():
previous_device = get_device()
from uuid import uuid4
# TestDevice1 supports arbitrary schedules, TestDevice2 does not
class TestDevice1(Device):
# These functions are needed during the setup of the defaultclock
def get_value(self, var):
return np.array([0.0001])
def add_array(self, var):
pass
def init_with_zeros(self, var, dtype):
pass
def fill_with_array(self, var, arr):
pass
class TestDevice2(TestDevice1):
def __init__(self):
super(TestDevice2, self).__init__()
self.network_schedule = ['start', 'groups', 'synapses',
'thresholds', 'resets', 'end']
# Unique names are important for getting the warnings again for multiple
# runs of the test suite
name1 = 'testdevice_' + str(uuid4())
name2 = 'testdevice_' + str(uuid4())
all_devices[name1] = TestDevice1()
all_devices[name2] = TestDevice2()
set_device(name1)
assert schedule_propagation_offset() == 0*ms
net = Network()
assert schedule_propagation_offset(net) == 0*ms
# Any schedule should work
net.schedule = list(reversed(net.schedule))
with catch_logs() as l:
net.run(0*ms)
assert len(l) == 0, 'did not expect a warning'
assert schedule_propagation_offset(net) == defaultclock.dt
set_device(name2)
assert schedule_propagation_offset() == defaultclock.dt
# Using the correct schedule should work
net.schedule = ['start', 'groups', 'synapses', 'thresholds', 'resets', 'end']
with catch_logs() as l:
net.run(0*ms)
assert len(l) == 0, 'did not expect a warning'
assert schedule_propagation_offset(net) == defaultclock.dt
# Using another (e.g. the default) schedule should raise a warning
net.schedule = None
with catch_logs() as l:
net.run(0*ms)
assert len(l) == 1 and l[0][1].endswith('schedule_conflict')
reset_device(previous_device)
def test_schedule_warning():
previous_device = get_device()
from uuid import uuid4
# TestDevice1 supports arbitrary schedules, TestDevice2 does not
class TestDevice1(Device):
# These functions are needed during the setup of the defaultclock
def get_value(self, var):
return np.array([0.0001])
def add_array(self, var):
pass
def init_with_zeros(self, var, dtype):
pass
def fill_with_array(self, var, arr):
pass
class TestDevice2(TestDevice1):
def __init__(self):
super(TestDevice2, self).__init__()
self.network_schedule = [
'start', 'groups', 'synapses', 'thresholds', 'resets', 'end'
]
# Unique names are important for getting the warnings again for multiple
# runs of the test suite
name1 = 'testdevice_' + str(uuid4())
name2 = 'testdevice_' + str(uuid4())
all_devices[name1] = TestDevice1()
all_devices[name2] = TestDevice2()
set_device(name1)
net = Network()
# Any schedule should work
net.schedule = list(reversed(net.schedule))
with catch_logs() as l:
net.run(0 * ms)
assert len(l) == 0, 'did not expect a warning'
set_device(name2)
# Using the correct schedule should work
net.schedule = [
'start', 'groups', 'synapses', 'thresholds', 'resets', 'end'
]
with catch_logs() as l:
net.run(0 * ms)
assert len(l) == 0, 'did not expect a warning'
# Using another (e.g. the default) schedule should raise a warning
net.schedule = None
with catch_logs() as l:
net.run(0 * ms)
assert len(l) == 1 and l[0][1].endswith('schedule_conflict')
reset_device(previous_device)
def test_resolution_warnings():
'''
Test that certain calls to resolve generate a warning.
'''
I = 3 * mV
tau = 5 * ms
another_I = 5 * mV
# Only specifying part of the namespace
expr = Expression('-(v + I) / tau', namespace={'I' : another_I},
exhaustive=False)
# make sure this triggers a warning (the 'I' in the namespace shadows the
# I variable defined above
with catch_logs() as logs:
namespace = expr.resolve(['v'])
assert len(logs) == 1
assert logs[0][0] == 'WARNING'
assert logs[0][1].endswith('resolution_conflict')
assert namespace['I'] == another_I and namespace['tau'] == tau
freq = 300 * Hz
t = 5 * second
# This expression treats t as a special variable and is not actually using
# the t above!
expr = Expression('sin(2 * 3.141 * freq * t)')
with catch_logs() as logs:
namespace = expr.resolve([])
assert len(logs) == 1
assert logs[0][0] == 'WARNING'
assert logs[0][1].endswith('resolution_conflict')
assert namespace['freq'] == freq and not 't' in namespace
I = 3 * mV
tau = 5 * ms
expr = Expression('-(v + I)/ tau')
# If we claim that I is an internal variable, it shadows the variable
# defined in the local namespace -- this should trigger a warning
with catch_logs() as logs:
namespace = expr.resolve(['v', 'I'])
assert len(logs) == 1
assert logs[0][0] == 'WARNING'
assert logs[0][1].endswith('resolution_conflict')
assert namespace['tau'] == tau and not 'I' in namespace
# A more extreme example: I is defined above, but also in the namespace and
# is claimed to be an internal variable
expr = Expression('-(v + I)/ tau', namespace={'I': 5 * mV},
exhaustive=False)
with catch_logs() as logs:
namespace = expr.resolve(['v', 'I'])
assert len(logs) == 1
assert logs[0][0] == 'WARNING'
assert logs[0][1].endswith('resolution_conflict')
assert namespace['tau'] == tau and not 'I' in namespace
def test_warning_internal_variables():
group1 = SimpleGroup(namespace=None,
variables={'N': Constant('N', Unit(1), 5)})
group2 = SimpleGroup(namespace=None,
variables={'N': Constant('N', Unit(1), 7)})
with catch_logs() as l:
group1.resolve_all(['N'], run_namespace={'N': 5}) # should not raise a warning
assert len(l) == 0, 'got warnings: %s' % str(l)
with catch_logs() as l:
group2.resolve_all(['N'], run_namespace={'N': 5}) # should raise a warning
assert len(l) == 1, 'got warnings: %s' % str(l)
assert l[0][1].endswith('.resolution_conflict')
def test_warning_internal_variables():
group1 = SimpleGroup(namespace=None,
variables={'N': Constant('N', 5)})
group2 = SimpleGroup(namespace=None,
variables={'N': Constant('N', 7)})
with catch_logs() as l:
group1.resolve_all(['N'], run_namespace={'N': 5}) # should not raise a warning
assert len(l) == 0, 'got warnings: %s' % str(l)
with catch_logs() as l:
group2.resolve_all(['N'], run_namespace={'N': 5}) # should raise a warning
assert len(l) == 1, 'got warnings: %s' % str(l)
assert l[0][1].endswith('.resolution_conflict')
def test_explicit_namespace():
''' Test resolution with an explicitly provided namespace '''
group = SimpleGroup(namespace={'variable': 42}, variables={})
# Explicitly provided
with catch_logs() as l:
assert group._resolve('variable', {}).get_value_with_unit() == 42
assert len(l) == 0
# Value from an explicit run namespace
with catch_logs() as l:
assert group._resolve('yet_another_var',
run_namespace={'yet_another_var': 17}).get_value_with_unit() == 17
assert len(l) == 0
def test_explicit_namespace():
''' Test resolution with an explicitly provided namespace '''
group = SimpleGroup(namespace={'variable': 42}, variables={})
# Explicitly provided
with catch_logs() as l:
assert group._resolve('variable', {}).get_value_with_unit() == 42
assert len(l) == 0
# Value from an explicit run namespace
with catch_logs() as l:
assert group._resolve('yet_another_var',
run_namespace={'yet_another_var': 17}).get_value_with_unit() == 17
assert len(l) == 0
def test_semantics_floor_division():
# See Brian2 github issues #815 and #661
G = NeuronGroup(11,
'''a : integer
b : integer
x : 1
y : 1
fvalue : 1
ivalue : integer''',
dtype={
'a': np.int32,
'b': np.int64,
'x': np.float,
'y': np.double
})
int_values = np.arange(-5, 6)
float_values = np.arange(-5.0, 6.0, dtype=np.double)
G.ivalue = int_values
G.fvalue = float_values
with catch_logs() as l:
G.run_regularly('''
a = ivalue//3
b = ivalue//3
x = fvalue//3
y = fvalue//3
''')
run(defaultclock.dt)
# XXX: brian2 test adapted here for 1 warning
assert len(l) == 1
assert_equal(G.a[:], int_values // 3)
assert_equal(G.b[:], int_values // 3)
assert_allclose(G.x[:], float_values // 3)
assert_allclose(G.y[:], float_values // 3)
def test_math_functions(func):
'''
Test that math functions give the same result, regardless of whether used
directly or in generated Python or C++ code.
'''
test_array = np.array([-1, -0.5, 0, 0.5, 1])
with catch_logs() as _: # Let's suppress warnings about illegal values
# 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(defaultclock.dt)
assert_allclose(
numpy_result,
mon.func_.flatten(),
err_msg='Function %s did not return the correct values' %
func.__name__)
def test_math_functions_short():
'''
Test that math functions give the same result, regardless of whether used
directly or in generated Python or C++ code. Test only a few functions
'''
default_dt = defaultclock.dt
test_array = np.array([-1, -0.5, 0, 0.5, 1])
with catch_logs() as _: # Let's suppress warnings about illegal values
# Functions with a single argument
for func in [exp, np.sqrt]:
# 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
# TODO: We are not testing the modulo operator here since it does
# not work for double values in C
for func, operator in [(np.power, '**')]:
# 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_missing_refractory_warning():
# Forgotten refractory argument
with catch_logs() as l:
group = NeuronGroup(1, 'dv/dt = -v / (10*ms) : 1 (unless refractory)',
threshold='v > 1', reset='v = 0')
assert len(l) == 1
assert l[0][0] == 'WARNING' and l[0][1].endswith('no_refractory')
def test_check_for_invalid_values_linear_integrator():
# A differential equation that cannot be solved by the linear
# integrator should return nan values to warn the user, and not silently
# return incorrect values. See discussion on
# https://github.com/brian-team/brian2/issues/626
a = 0.0 / ms
b = 1.0 / ms
c = -0.5 / ms
d = -0.1 / ms
eqs = '''
dx/dt = a * x + b * y : 1
dy/dt = c * x + d * y : 1
'''
G = NeuronGroup(1, eqs, threshold='x > 100', reset='x = 0', method='exact')
G.x = 1
BrianLogger._log_messages.clear(
) # because the log message is set to be shown only once
with catch_logs() as clog:
try:
run(1 * ms)
# this check allows for the possibility that we improve the linear
# integrator in the future so that it can handle this equation
if numpy.isnan(G.x[0]):
assert 'invalid_values' in repr(clog)
else:
assert G.x[0] != 0
except UnsupportedEquationsException:
pass
def test_missing_refractory_warning():
# Forgotten refractory argument
with catch_logs() as l:
group = NeuronGroup(1, 'dv/dt = -v / (10*ms) : 1 (unless refractory)',
threshold='v > 1', reset='v = 0')
assert len(l) == 1
assert l[0][0] == 'WARNING' and l[0][1].endswith('no_refractory')
def test_compiler_error():
# In particular on OSX with clang in a conda environment, compilation might fail.
# Switching to a system gcc might help in such cases. Make sure that the error
# message mentions that.
old_CC = os.environ.get('CC', None)
old_CXX = os.environ.get('CXX', None)
os.environ.update({
'CC': 'non-existing-compiler',
'CXX': 'non-existing-compiler++'
})
try:
with catch_logs() as l:
assert not CythonCodeObject.is_available()
assert len(l) > 0 # There are additional warnings about compiler flags
last_warning = l[-1]
assert last_warning[1].endswith('.failed_compile_test')
assert 'CC' in last_warning[2] and 'CXX' in last_warning[2]
finally:
if old_CC:
os.environ['CC'] = old_CC
else:
del os.environ['CC']
if old_CXX:
os.environ['CXX'] = old_CXX
else:
del os.environ['CXX']
def test_math_functions(func, needs_c99_support):
'''
Test that math functions give the same result, regardless of whether used
directly or in generated Python or C++ code.
'''
if not get_device() == RuntimeDevice or prefs.codegen.target != 'numpy':
if needs_c99_support and not compiler_supports_c99():
pytest.skip('Support for function "{}" needs a compiler with C99 '
'support.')
test_array = np.array([-1, -0.5, 0, 0.5, 1])
with catch_logs() as _: # Let's suppress warnings about illegal values
# 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(defaultclock.dt)
assert_allclose(
numpy_result,
mon.func_.flatten(),
err_msg='Function %s did not return the correct values' %
func.__name__)
def test_transmission_scalar_delay_different_clocks():
inp = SpikeGeneratorGroup(
2,
[0, 1],
[0, 1] * ms,
dt=0.5 * ms,
# give the group a unique name to always
# get a 'fresh' warning
name='sg_%d' % uuid.uuid4())
target = NeuronGroup(2, 'v:1', dt=0.1 * ms)
S = Synapses(inp, target, pre='v+=1', delay=0.5 * ms, connect='i==j')
mon = StateMonitor(target, 'v', record=True)
net = Network(inp, target, S, mon)
# We should get a warning when using inconsistent dts
with catch_logs() as l:
net.run(2 * ms)
assert len(l) == 1, 'expected a warning, got %d' % len(l)
assert l[0][1].endswith('synapses_dt_mismatch')
assert_equal(mon[0].v[mon.t < 0.5 * ms], 0)
assert_equal(mon[0].v[mon.t >= 0.5 * ms], 1)
assert_equal(mon[1].v[mon.t < 1.5 * ms], 0)
assert_equal(mon[1].v[mon.t >= 1.5 * ms], 1)
def test_dependency_check():
def create_net():
G = NeuronGroup(10, 'v: 1', threshold='False')
dependent_objects = [
StateMonitor(G, 'v', record=True),
SpikeMonitor(G),
PopulationRateMonitor(G),
Synapses(G, G, pre='v+=1', connect=True)
]
return dependent_objects
dependent_objects = create_net()
# Trying to simulate the monitors/synapses without the group should fail
for obj in dependent_objects:
assert_raises(ValueError, lambda: Network(obj).run(0*ms))
# simulation with a magic network should work when we have an explicit
# reference to one of the objects, but the object should be inactive and
# we should get a warning
assert all(obj.active for obj in dependent_objects)
for obj in dependent_objects: # obj is our explicit reference
with catch_logs() as l:
run(0*ms)
dependency_warnings = [msg[2] for msg in l
if msg[1] == 'brian2.core.magic.dependency_warning']
assert len(dependency_warnings) == 1
assert not obj.active
def test_synapses_access_subgroups_problematic():
G1 = NeuronGroup(5, 'x:1')
G2 = NeuronGroup(10, 'y:1')
SG1 = G1[2:5]
SG2 = G2[4:9]
S = Synapses(G1, G2, 'w:1')
S.connect()
# Note that "j" is not ambiguous, because the equivalent in the target group
# is called "i" (this previously raised a warning)
tests = [
((SG1, slice(None)), 'i', 1),
((SG1, slice(None)), 'i + N_pre', 2),
((SG1, slice(None)), 'N_pre', 1),
((slice(None), SG2), 'j', 0),
((slice(None), SG2), 'N_post', 1),
((slice(None), SG2), 'N', 1),
((SG1, SG2), 'i', 1),
((SG1, SG2), 'i + j', 1),
((SG1, SG2), 'N_pre', 1),
((SG1, SG2), 'j', 0),
((SG1, SG2), 'N_post', 1),
((SG1, SG2), 'N', 1),
# These should not raise a warning
((SG1, SG2), 'w', 0),
((SG1, SG2), 'x_pre', 0),
((SG1, SG2), 'y_post', 0),
((SG1, SG2), 'y', 0)
]
for item, value, n_warnings in tests:
with catch_logs() as l:
S.w.__setitem__(item, value)
assert len(l) == n_warnings, 'expected %d, got %d warnings' % (n_warnings, len(l))
assert all([entry[1].endswith('ambiguous_string_expression')
for entry in l])
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_multiple_noise_variables_deterministic_noise():
# The "random" values are always 0.5
@implementation('cpp',
'''
double randn(int vectorisation_idx)
{
return 0.5;
}
''')
@implementation('cython',
'''
cdef double randn(int vectorisation_idx):
return 0.5
''')
@check_units(N=Unit(1), result=Unit(1))
def fake_randn(N):
return 0.5*ones(N)
old_randn = DEFAULT_FUNCTIONS['randn']
DEFAULT_FUNCTIONS['randn'] = fake_randn
all_eqs = ['''dx/dt = y : 1
dy/dt = -y / (10*ms) + dt**-.5*0.5*ms**-1.5 + dt**-.5*0.5*ms**-1.5: Hz
''',
'''dx/dt = y + dt**-.5*0.5*ms**-0.5: 1
dy/dt = -y / (10*ms) + dt**-.5*0.5 * ms**-1.5 : Hz
''']
all_eqs_noise = ['''dx/dt = y : 1
dy/dt = -y / (10*ms) + xi_1 * ms**-1.5 + xi_2 * ms**-1.5: Hz
''',
'''dx/dt = y + xi_1*ms**-0.5: 1
dy/dt = -y / (10*ms) + xi_2 * ms**-1.5 : Hz
''']
for eqs, eqs_noise in zip(all_eqs, all_eqs_noise):
G = NeuronGroup(2, eqs, method='euler')
G.x = [5, 17]
G.y = [25, 5 ] * Hz
mon = StateMonitor(G, ['x', 'y'], record=True)
net = Network(G, mon)
net.run(10*ms)
no_noise_x, no_noise_y = mon.x[:], mon.y[:]
for method_name, method in [('euler', euler), ('milstein', milstein)]:
# Note that for milstein, the check for diagonal noise will fail, but
# it should still work since the two noise variables really do only
# present a single variable
with catch_logs('WARNING'):
G = NeuronGroup(2, eqs_noise, method=method)
G.x = [5, 17]
G.y = [25, 5 ] * Hz
mon = StateMonitor(G, ['x', 'y'], record=True)
net = Network(G, mon)
# We run it deterministically, but still we'd detect major errors (e.g.
# non-stochastic terms that are added twice, see #330
net.run(10*ms, namespace={'noise_factor': 0})
assert_allclose(mon.x[:], no_noise_x,
err_msg='Method %s gave incorrect results' % method_name)
assert_allclose(mon.y[:], no_noise_y,
err_msg='Method %s gave incorrect results' % method_name)
def test_explicit_namespace():
''' Test resolution with an explicitly provided namespace '''
explicit_namespace = {'variable': 'explicit_var',
'randn': 'explicit_randn'}
# Explicitly provided
namespace = create_namespace( explicit_namespace)
with catch_logs() as l:
assert namespace['variable'] == 'explicit_var'
assert len(l) == 0
with catch_logs() as l:
# The explicitly provided namespace should not overwrite functions
assert isinstance(namespace['randn'], RandnFunction)
_assert_one_warning(l)
def test_warning():
from brian2.core.functions import DEFAULT_FUNCTIONS
from brian2.units.stdunits import cm as brian_cm
# Name in external namespace clashes with unit/function name
exp = 23
cm = 42
group = SimpleGroup(namespace=None, variables={})
with catch_logs() as l:
resolved = group.resolve_all(['exp'])['exp']
assert resolved == DEFAULT_FUNCTIONS['exp']
assert len(l) == 1, 'got warnings: %s' % str(l)
assert l[0][1].endswith('.resolution_conflict')
with catch_logs() as l:
resolved = group.resolve_all(['cm'])['cm']
assert resolved.get_value_with_unit() == brian_cm
assert len(l) == 1, 'got warnings: %s' % str(l)
assert l[0][1].endswith('.resolution_conflict')
def test_warning():
from brian2.core.functions import DEFAULT_FUNCTIONS
from brian2.units.stdunits import cm as brian_cm
# Name in external namespace clashes with unit/function name
exp = 23
cm = 42
group = SimpleGroup(namespace=None, variables={})
with catch_logs() as l:
resolved = group.resolve_all(['exp'])['exp']
assert resolved == DEFAULT_FUNCTIONS['exp']
assert len(l) == 1, 'got warnings: %s' % str(l)
assert l[0][1].endswith('.resolution_conflict')
with catch_logs() as l:
resolved = group.resolve_all(['cm'])['cm']
assert resolved.get_value_with_unit() == brian_cm
assert len(l) == 1, 'got warnings: %s' % str(l)
assert l[0][1].endswith('.resolution_conflict')
def test_warning():
from brian2.core.functions import DEFAULT_FUNCTIONS
from brian2.units.stdunits import cm as brian_cm
# Name in external namespace clashes with unit/function name
exp = 23
cm = 42
namespace = create_namespace()
local_ns = get_local_namespace(level=0)
with catch_logs() as l:
resolved = namespace.resolve('exp', ('implicit namespace', local_ns))
assert resolved == DEFAULT_FUNCTIONS['exp']
assert len(l) == 1
assert l[0][1].endswith('.resolution_conflict')
with catch_logs() as l:
resolved = namespace.resolve('cm', ('implicit namespace', local_ns))
assert resolved == brian_cm
assert len(l) == 1
assert l[0][1].endswith('.resolution_conflict')
def test_namespace_warnings():
G = NeuronGroup(1, '''x : 1
y : 1''')
# conflicting variable in namespace
y = 5
with catch_logs() as l:
G.x = 'y'
assert len(l) == 1
assert l[0][1].endswith('.resolution_conflict')
# conflicting variables with special meaning
i = 5
N = 3
with catch_logs() as l:
G.x = 'i / N'
assert len(l) == 2
assert l[0][1].endswith('.resolution_conflict')
assert l[1][1].endswith('.resolution_conflict')
def test_delete_directory():
set_device('cpp_standalone', build_on_run=True, directory=None)
group = NeuronGroup(10, 'dv/dt = -v / (10*ms) : volt', method='exact')
group.v = np.arange(10) * mV # uses the static array mechanism
run(defaultclock.dt)
# Add a new file
dummy_file = os.path.join(device.project_dir, 'results', 'dummy.txt')
open(dummy_file, 'w').flush()
assert os.path.isfile(dummy_file)
with catch_logs() as logs:
device.delete(directory=True)
assert len(logs) == 1
assert os.path.isfile(dummy_file)
with catch_logs() as logs:
device.delete(directory=True, force=True)
assert len(logs) == 0
# everything should be deleted
assert not os.path.exists(device.project_dir)
def test_explicit_namespace():
''' Test resolution with an explicitly provided namespace '''
explicit_namespace = {
'variable': 'explicit_var',
'randn': 'explicit_randn'
}
# Explicitly provided
namespace = create_namespace(explicit_namespace)
with catch_logs() as l:
assert namespace['variable'] == 'explicit_var'
assert len(l) == 0
with catch_logs() as l:
# The explicitly provided namespace should not overwrite functions
assert isinstance(namespace['randn'], RandnFunction)
_assert_one_warning(l)
def test_no_synapses():
# Synaptic pathway but no synapses
G1 = NeuronGroup(1, '', threshold='True')
G2 = NeuronGroup(1, 'v:1')
S = Synapses(G1, G2, pre='v+=1', name='synapses_'+str(uuid.uuid4()).replace('-', '_'))
net = Network(G1, G2, S)
with catch_logs() as l:
net.run(defaultclock.dt)
assert len(l) == 1, 'expected 1 warning, got %d' % len(l)
assert l[0][1].endswith('.no_synapses')
def test_explicit_namespace():
''' Test resolution with an explicitly provided namespace '''
explicit_namespace = {'variable': 'explicit_var'}
# Explicitly provided
namespace = create_namespace( explicit_namespace)
with catch_logs() as l:
assert namespace['variable'] == 'explicit_var'
assert len(l) == 0
def test_multiple_noise_variables_deterministic_noise(
fake_randn_randn_fixture):
all_eqs = [
'''dx/dt = y : 1
dy/dt = -y / (10*ms) + dt**-.5*0.5*ms**-1.5 + dt**-.5*0.5*ms**-1.5: Hz
''', '''dx/dt = y + dt**-.5*0.5*ms**-0.5: 1
dy/dt = -y / (10*ms) + dt**-.5*0.5 * ms**-1.5 : Hz
'''
]
all_eqs_noise = [
'''dx/dt = y : 1
dy/dt = -y / (10*ms) + xi_1 * ms**-1.5 + xi_2 * ms**-1.5: Hz
''', '''dx/dt = y + xi_1*ms**-0.5: 1
dy/dt = -y / (10*ms) + xi_2 * ms**-1.5 : Hz
'''
]
monitors_no_noise = []
monitors = []
for eqs, eqs_noise in zip(all_eqs, all_eqs_noise):
G = NeuronGroup(2, eqs, method='euler')
G.x = [5, 17]
G.y = [25, 5] * Hz
mon = StateMonitor(G, ['x', 'y'], record=True)
net = Network(G, mon)
net.run(10 * ms)
monitors_no_noise.append(mon)
monitors.append(dict())
for method_name, method in [('euler', euler), ('heun', heun)]:
with catch_logs('WARNING'):
G = NeuronGroup(2, eqs_noise, method=method)
G.x = [5, 17]
G.y = [25, 5] * Hz
mon = StateMonitor(G, ['x', 'y'], record=True)
net = Network(G, mon)
net.run(10 * ms)
monitors[-1][method_name] = mon
device.build(direct_call=False, **device.build_options)
for i in range(len(all_eqs)):
mon = monitors_no_noise[i]
no_noise_x, no_noise_y = mon.x[:], mon.y[:]
for method_name in ['euler', 'heun']:
mon = monitors[i][method_name]
assert_allclose(
mon.x[:],
no_noise_x,
err_msg=f'Method {method_name} gave incorrect results')
assert_allclose(
mon.y[:],
no_noise_y,
err_msg=f'Method {method_name} gave incorrect results')
def test_multiple_noise_variables_deterministic_noise():
# The "random" values are always 0.5
@implementation('cpp',
'''
double randn(int vectorisation_idx)
{
return 0.5;
}
''')
@implementation('cython',
'''
cdef double randn(int vectorisation_idx):
return 0.5
''')
@check_units(N=Unit(1), result=Unit(1))
def fake_randn(N):
return 0.5*ones(N)
old_randn = DEFAULT_FUNCTIONS['randn']
DEFAULT_FUNCTIONS['randn'] = fake_randn
all_eqs = ['''dx/dt = y : 1
dy/dt = -y / (10*ms) + dt**-.5*0.5*ms**-1.5 + dt**-.5*0.5*ms**-1.5: Hz
''',
'''dx/dt = y + dt**-.5*0.5*ms**-0.5: 1
dy/dt = -y / (10*ms) + dt**-.5*0.5 * ms**-1.5 : Hz
''']
all_eqs_noise = ['''dx/dt = y : 1
dy/dt = -y / (10*ms) + xi_1 * ms**-1.5 + xi_2 * ms**-1.5: Hz
''',
'''dx/dt = y + xi_1*ms**-0.5: 1
dy/dt = -y / (10*ms) + xi_2 * ms**-1.5 : Hz
''']
for eqs, eqs_noise in zip(all_eqs, all_eqs_noise):
G = NeuronGroup(2, eqs, method='euler')
G.x = [5, 17]
G.y = [25, 5 ] * Hz
mon = StateMonitor(G, ['x', 'y'], record=True)
net = Network(G, mon)
net.run(10*ms)
no_noise_x, no_noise_y = mon.x[:], mon.y[:]
for method_name, method in [('euler', euler), ('heun', heun)]:
with catch_logs('WARNING'):
G = NeuronGroup(2, eqs_noise, method=method)
G.x = [5, 17]
G.y = [25, 5 ] * Hz
mon = StateMonitor(G, ['x', 'y'], record=True)
net = Network(G, mon)
net.run(10*ms)
assert_allclose(mon.x[:], no_noise_x,
err_msg='Method %s gave incorrect results' % method_name)
assert_allclose(mon.y[:], no_noise_y,
err_msg='Method %s gave incorrect results' % method_name)
def test_math_functions_short():
'''
Test that math functions give the same result, regardless of whether used
directly or in generated Python or C++ code. Test only a few functions
'''
default_dt = defaultclock.dt
test_array = np.array([-1, -0.5, 0, 0.5, 1])
with catch_logs() as _: # Let's suppress warnings about illegal values
# Functions with a single argument
for func in [exp, np.sqrt]:
# 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
# TODO: We are not testing the modulo operator here since it does
# not work for double values in C
for func, operator in [(np.power, '**')]:
# 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_schedule_warning():
previous_device = get_device()
from uuid import uuid4
# TestDevice1 supports arbitrary schedules, TestDevice2 does not
class TestDevice1(Device):
pass
class TestDevice2(Device):
def __init__(self):
super(TestDevice2, self).__init__()
self.network_schedule = ['start', 'groups', 'synapses',
'thresholds', 'resets', 'end']
# Unique names are important for getting the warnings again for multiple
# runs of the test suite
name1 = 'testdevice_' + str(uuid4())
name2 = 'testdevice_' + str(uuid4())
all_devices[name1] = TestDevice1()
all_devices[name2] = TestDevice2()
set_device(name1)
net = Network()
# Any schedule should work
net.schedule = list(reversed(net.schedule))
with catch_logs() as l:
net.run(0*ms)
assert len(l) == 0, 'did not expect a warning'
set_device(name2)
# Using the correct schedule should work
net.schedule = ['start', 'groups', 'synapses', 'thresholds', 'resets', 'end']
with catch_logs() as l:
net.run(0*ms)
assert len(l) == 0, 'did not expect a warning'
# Using another (e.g. the default) schedule should raise a warning
net.schedule = None
with catch_logs() as l:
net.run(0*ms)
assert len(l) == 1 and l[0][1].endswith('schedule_conflict')
set_device(previous_device)
def test_state_variables_group_as_index_problematic():
G = NeuronGroup(10, 'v : 1')
SG = G[4:9]
G.v = 1
tests = [('i', 1), ('N', 1), ('N + i', 2), ('v', 0)]
for value, n_warnings in tests:
with catch_logs() as l:
G.v.__setitem__(SG, value)
assert len(l) == n_warnings, 'expected %d, got %d warnings' % (
n_warnings, len(l))
assert all([
entry[1].endswith('ambiguous_string_expression') for entry in l
])
def test_no_synapses():
# Synaptic pathway but no synapses
G1 = NeuronGroup(1, '', threshold='True')
G2 = NeuronGroup(1, 'v:1')
S = Synapses(G1,
G2,
pre='v+=1',
name='synapses_' + str(uuid.uuid4()).replace('-', '_'))
net = Network(G1, G2, S)
with catch_logs() as l:
net.run(defaultclock.dt)
assert len(l) == 1, 'expected 1 warning, got %d' % len(l)
assert l[0][1].endswith('.no_synapses')
def test_poissongroup_namespace():
rate_const = 0*Hz
P = PoissonGroup(1, rates='rate_const', namespace={'rate_const':
1/defaultclock.dt},
name='poissongroup_%s' % (uuid.uuid4().hex))
P2 = PoissonGroup(1, rates='rate_const')
mon = SpikeMonitor(P)
mon2 = SpikeMonitor(P2)
with catch_logs() as l:
run(2*defaultclock.dt)
assert len(l) == 1
assert l[0][1].endswith('resolution_conflict')
assert mon.num_spikes == 2
assert mon2.num_spikes == 0
def test_multiple_noise_variables_extended():
# Some actual simulations with multiple noise variables
eqs = '''dx/dt = y : 1
dy/dt = - 1*ms**-1*y - 40*ms**-2*x : Hz
'''
all_eqs_noise = [
'''dx/dt = y : 1
dy/dt = noise_factor*ms**-1.5*xi_1 + noise_factor*ms**-1.5*xi_2
- 1*ms**-1*y - 40*ms**-2*x : Hz
''', '''dx/dt = y + noise_factor*ms**-0.5*xi_1: 1
dy/dt = noise_factor*ms**-1.5*xi_2
- 1*ms**-1*y - 40*ms**-2*x : Hz
'''
]
G = NeuronGroup(2, eqs, method='euler')
G.x = [0.5, 1]
G.y = [0, 0.5] * Hz
mon1 = StateMonitor(G, ['x', 'y'], record=True)
net = Network(G, mon1)
net.run(10 * ms)
monitors = []
for eqs_noise in all_eqs_noise:
monitors.append(dict())
for method_name, method in [('euler', euler), ('heun', heun)]:
with catch_logs('WARNING'):
G = NeuronGroup(2, eqs_noise, method=method)
G.x = [0.5, 1]
G.y = [0, 0.5] * Hz
mon = StateMonitor(G, ['x', 'y'], record=True)
net = Network(G, mon)
# We run it deterministically, but still we'd detect major errors (e.g.
# non-stochastic terms that are added twice, see #330
net.run(10 * ms, namespace={'noise_factor': 0})
monitors[-1][method_name] = mon
device.build(direct_call=False, **device.build_options)
no_noise_x, no_noise_y = mon1.x[:], mon1.y[:]
for i in range(len(all_eqs_noise)):
for method_name in ['euler', 'heun']:
mon = monitors[i][method_name]
assert_allclose(
mon.x[:],
no_noise_x,
err_msg=f'Method {method_name} gave incorrect results')
assert_allclose(
mon.y[:],
no_noise_y,
err_msg=f'Method {method_name} gave incorrect results')
def test_poissongroup_namespace():
rate_const = 0*Hz
P = PoissonGroup(1, rates='rate_const', namespace={'rate_const':
1/defaultclock.dt},
name='poissongroup_%s' % (uuid.uuid4().hex))
P2 = PoissonGroup(1, rates='rate_const')
mon = SpikeMonitor(P)
mon2 = SpikeMonitor(P2)
with catch_logs() as l:
run(2*defaultclock.dt)
assert len(l) == 1
assert l[0][1].endswith('resolution_conflict')
assert mon.num_spikes == 2
assert mon2.num_spikes == 0
def test_state_variables_group_as_index_problematic():
G = NeuronGroup(10, 'v : 1')
SG = G[4:9]
G.v = 1
tests = [('i', 1),
('N', 1),
('N + i', 2),
('v', 0)]
for value, n_warnings in tests:
with catch_logs() as l:
G.v.__setitem__(SG, value)
assert len(l) == n_warnings, 'expected %d, got %d warnings' % (n_warnings, len(l))
assert all([entry[1].endswith('ambiguous_string_expression')
for entry in l])
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_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_spikegenerator_multiple_runs():
indices = np.zeros(5)
times = np.arange(5) * ms
spike_gen = SpikeGeneratorGroup(1, indices, times) # all good
spike_mon = SpikeMonitor(spike_gen)
run(5 * ms)
# Setting the same spike times again should not do anything, since they are
# before the start of the current simulation
spike_gen.set_spikes(indices, times)
# however, a warning should be raised
with catch_logs() as l:
run(5 * ms)
device.build(direct_call=False, **device.build_options)
assert len(l) == 1 and l[0][1].endswith('ignored_spikes')
assert spike_mon.num_spikes == 5
def test_spikegenerator_multiple_runs():
indices = np.zeros(5)
times = np.arange(5)*ms
spike_gen = SpikeGeneratorGroup(1, indices, times) # all good
spike_mon = SpikeMonitor(spike_gen)
run(5*ms)
# Setting the same spike times again should not do anything, since they are
# before the start of the current simulation
spike_gen.set_spikes(indices, times)
# however, a warning should be raised
with catch_logs() as l:
run(5*ms)
device.build(direct_call=False, **device.build_options)
assert len(l) == 1 and l[0][1].endswith('ignored_spikes')
assert spike_mon.num_spikes == 5
def test_profile_warning():
# Test that profiling info is not 0
set_device('cuda_standalone', directory=None)
prefs.devices.cuda_standalone.profile = True
G = NeuronGroup(1, 'v:1', threshold='True')
BrianLogger._log_messages.clear()
with catch_logs() as logs:
run(defaultclock.dt, profile=True)
assert len(logs) == 1, len(logs)
assert logs[0][0] == 'WARNING'
assert logs[0][1] == 'brian2.devices.cuda_standalone'
def test_magic_weak_reference():
'''
Test that holding a weak reference to an object does not make it get
simulated.'''
G1 = NeuronGroup(1, 'v:1')
# this object should not be included
G2 = weakref.ref(NeuronGroup(1, 'v:1'))
with catch_logs(log_level=logging.DEBUG) as l:
run(1*ms)
# Check the debug messages for the number of included objects
magic_objects = [msg[2] for msg in l
if msg[1] == 'brian2.core.magic.magic_objects'][0]
assert '2 objects' in magic_objects, 'Unexpected log message: %s' % magic_objects
def test_magic_unused_object():
'''Test that creating unused objects does not affect the magic system.'''
def create_group():
# Produce two objects but return only one
G1 = NeuronGroup(1, 'v:1') # no Thresholder or Resetter
G2 = NeuronGroup(1, 'v:1') # This object should be garbage collected
return G1
G = create_group()
with catch_logs(log_level=logging.INFO) as l:
run(1*ms)
# Check the debug messages for the number of included objects
magic_objects = [msg[2] for msg in l
if msg[1] == 'brian2.core.magic.magic_objects'][0]
assert '2 objects' in magic_objects, 'Unexpected log message: %s' % magic_objects
def test_transmission_scalar_delay_different_clocks():
inp = SpikeGeneratorGroup(2, [0, 1], [0, 1]*ms, dt=0.5*ms,
# give the group a unique name to always
# get a 'fresh' warning
name='sg_%d' % uuid.uuid4())
target = NeuronGroup(2, 'v:1', dt=0.1*ms)
S = Synapses(inp, target, pre='v+=1', delay=0.5*ms, connect='i==j')
mon = StateMonitor(target, 'v', record=True)
net = Network(inp, target, S, mon)
# We should get a warning when using inconsistent dts
with catch_logs() as l:
net.run(2*ms)
assert len(l) == 1, 'expected a warning, got %d' % len(l)
assert l[0][1].endswith('synapses_dt_mismatch')
assert_equal(mon[0].v[mon.t<0.5*ms], 0)
assert_equal(mon[0].v[mon.t>=0.5*ms], 1)
assert_equal(mon[1].v[mon.t<1.5*ms], 0)
assert_equal(mon[1].v[mon.t>=1.5*ms], 1)
def test_multiple_noise_variables_deterministic_noise():
DEFAULT_FUNCTIONS["randn"] = fake_randn
all_eqs = [
"""dx/dt = y : 1
dy/dt = -y / (10*ms) + dt**-.5*0.5*ms**-1.5 + dt**-.5*0.5*ms**-1.5: Hz
""",
"""dx/dt = y + dt**-.5*0.5*ms**-0.5: 1
dy/dt = -y / (10*ms) + dt**-.5*0.5 * ms**-1.5 : Hz
""",
]
all_eqs_noise = [
"""dx/dt = y : 1
dy/dt = -y / (10*ms) + xi_1 * ms**-1.5 + xi_2 * ms**-1.5: Hz
""",
"""dx/dt = y + xi_1*ms**-0.5: 1
dy/dt = -y / (10*ms) + xi_2 * ms**-1.5 : Hz
""",
]
for eqs, eqs_noise in zip(all_eqs, all_eqs_noise):
G = NeuronGroup(2, eqs, method="euler")
G.x = [5, 17]
G.y = [25, 5] * Hz
mon = StateMonitor(G, ["x", "y"], record=True)
net = Network(G, mon)
net.run(10 * ms)
no_noise_x, no_noise_y = mon.x[:], mon.y[:]
for method_name, method in [("euler", euler), ("heun", heun)]:
with catch_logs("WARNING"):
G = NeuronGroup(2, eqs_noise, method=method)
G.x = [5, 17]
G.y = [25, 5] * Hz
mon = StateMonitor(G, ["x", "y"], record=True)
net = Network(G, mon)
net.run(10 * ms)
assert_allclose(mon.x[:], no_noise_x, err_msg="Method %s gave incorrect results" % method_name)
assert_allclose(mon.y[:], no_noise_y, err_msg="Method %s gave incorrect results" % method_name)
def test_multiple_noise_variables_extended():
# Some actual simulations with multiple noise variables
eqs = '''dx/dt = y : 1
dy/dt = - 1*ms**-1*y - 40*ms**-2*x : Hz
'''
all_eqs_noise = ['''dx/dt = y : 1
dy/dt = noise_factor*ms**-1.5*xi_1 + noise_factor*ms**-1.5*xi_2
- 1*ms**-1*y - 40*ms**-2*x : Hz
''',
'''dx/dt = y + noise_factor*ms**-0.5*xi_1: 1
dy/dt = noise_factor*ms**-1.5*xi_2
- 1*ms**-1*y - 40*ms**-2*x : Hz
''']
G = NeuronGroup(2, eqs, method='euler')
G.x = [0.5, 1]
G.y = [0, 0.5] * Hz
mon = StateMonitor(G, ['x', 'y'], record=True)
net = Network(G, mon)
net.run(10*ms)
no_noise_x, no_noise_y = mon.x[:], mon.y[:]
for eqs_noise in all_eqs_noise:
for method_name, method in [('euler', euler), ('heun', heun), ('milstein', milstein)]:
# Note that for milstein, the check for diagonal noise will fail, but
# it should still work since the two noise variables really do only
# present a single variable
with catch_logs('WARNING'):
G = NeuronGroup(2, eqs_noise, method=method)
G.x = [0.5, 1]
G.y = [0, 0.5] * Hz
mon = StateMonitor(G, ['x', 'y'], record=True)
net = Network(G, mon)
# We run it deterministically, but still we'd detect major errors (e.g.
# non-stochastic terms that are added twice, see #330
net.run(10*ms, namespace={'noise_factor': 0})
assert_allclose(mon.x[:], no_noise_x,
err_msg='Method %s gave incorrect results' % method_name)
assert_allclose(mon.y[:], no_noise_y,
err_msg='Method %s gave incorrect results' % method_name)