def test_emergent_neuron(self):
"""Test quantitative equivalence with emergent on the neuron tutorial."""
check = True
for adapt_on in [False, True]:
neuron_data = data.parse_unit(
'neuron_adapt.dat' if adapt_on else 'neuron.dat')
spec = leabra.UnitSpec(adapt_on=adapt_on,
noisy_act=True,
g_bar_e=0.3,
g_bar_l=0.3,
g_bar_i=1.0,
act_gain=40,
act_sd=0.01)
log_names = ('net', 'I_net', 'v_m', 'act', 'v_m_eq', 'adapt',
'avg_ss', 'avg_s', 'avg_m', 'avg_s_eff', 'avg_l')
receiver = leabra.Unit(spec=spec, log_names=log_names)
inputs = 10 * [0.0] + 150 * [1.0] + 40 * [0.0]
for g_e in inputs:
receiver.add_excitatory(g_e)
receiver.calculate_net_in()
receiver.cycle('minus')
# note that here there seems to be a bit of drift on I_net after the 190 time mark.
# possibly because of the single/double float precision discrepancy.
check = quantitative_match(receiver.logs,
neuron_data,
rtol=2e-05,
atol=1e-08,
check=check)
self.assertTrue(check)
def test_act(self):
"""Test that the unit activity is behaving well under excitation."""
spec = leabra.UnitSpec(g_bar_e=0.40,
g_bar_l=2.80,
g_bar_i=1.00,
g_l=1.0,
e_rev_e=1.00,
e_rev_l=0.15,
e_rev_i=0.15,
act_thr=0.25,
act_gain=600.00,
act_sd=0.01)
u = leabra.Unit(spec=spec)
for _ in range(15):
u.add_excitatory(1.0)
u.calculate_net_in()
u.cycle('minus')
for _ in range(150):
u.add_excitatory(1.0)
u.calculate_net_in()
u.cycle('minus')
self.assertTrue(0.85 < u.act <= 0.95)
for _ in range(10):
u.add_excitatory(0.0)
u.calculate_net_in()
u.cycle('minus')
self.assertTrue(u.act < 0.05)
def test_emergent_neuron(self):
"""Test quantitative equivalence with emergent on the neuron tutorial."""
for adapt_on in [False, True]:
neuron_data = data.parse_unit(
'neuron_adapt.txt' if adapt_on else 'neuron.txt')
spec = leabra.UnitSpec(adapt_on=adapt_on, noisy_act=True)
log_names = ('net', 'I_net', 'v_m', 'act', 'v_m_eq', 'adapt',
'avg_ss', 'avg_s', 'avg_m', 'avg_s_eff', 'avg_l')
receiver = leabra.Unit(spec=spec, log_names=log_names)
inputs = 10 * [0.0] + 150 * [1.0] + 40 * [0.0]
for g_e in inputs:
receiver.add_excitatory(g_e)
receiver.calculate_net_in()
receiver.cycle()
check = True
for name in receiver.logs.keys():
for t, (py, em) in enumerate(
zip(receiver.logs[name], neuron_data[name])):
if not np.allclose(py, em, rtol=1e-05, atol=1e-07):
print(
'{}:{:03d} [py] {: .10f} != {: .10f} [emergent] ({}adapt)'
.format(name, t, py, em,
'' if adapt_on else 'no '))
check = False
elif name == 'I_net':
print(
'{}:{:03d} [py] {: .10f} == {: .10f} [emergent] ({}adapt)'
.format(name, t, py, em,
'' if adapt_on else 'no '))
self.assertTrue(check)
def test_spec(self):
"""Test the behavior of spec"""
u_spec = leabra.UnitSpec()
u_spec.act_thr = 0.25
u = leabra.Unit(spec=u_spec) # by default, copyspec=False
u2 = leabra.Unit(spec=u_spec)
self.assertEqual(u.spec.act_thr, 0.25)
self.assertEqual(u2.spec.act_thr, 0.25)
u.spec.act_thr = 0.30
self.assertEqual(u.spec.act_thr, 0.30)
self.assertEqual(u2.spec.act_thr, 0.30)
u3 = leabra.Unit(spec=u_spec.copy()) # u3 has an independent spec
u3.spec.act_thr = 0.45
self.assertEqual(u2.spec.act_thr, 0.30)
self.assertEqual(u3.spec.act_thr, 0.45)
def test_xx1_thr(self):
"""Test the threshold for xx1 functions"""
u_spec = leabra.UnitSpec()
u_spec.act_thr = 0.25
u = leabra.Unit(spec=u_spec)
self.assertEqual(u_spec.xx1(-0.1), 0.0)
self.assertTrue(0.0 < u_spec.xx1(0.1))
self.assertEqual(u_spec.noisy_xx1(-0.1), 0.0)
self.assertTrue(0.1 < u_spec.noisy_xx1(0.1))
def test_avgs_forced(self):
"""Test if units with forced activity update their averages"""
u = leabra.Unit()
for t in range(10):
u.force_activity(1.0)
u.calculate_net_in()
u.cycle('minus')
for name in ['avg_ss', 'avg_s', 'avg_m', 'avg_s_eff']:
self.assertTrue(getattr(u, name) != 0.15)
def test_avg_l(self):
"""Test that the long-term average are correctly updated."""
u = leabra.Unit()
for _ in range(20):
u.add_excitatory(1.0)
u.calculate_net_in()
u.cycle()
self.assertEqual(u.avg_l, 0.40)
u.spec.update_avg_l(u)
self.assertTrue(np.allclose(0.52, u.avg_l, rtol=0.1, atol=0.1))
#TODO: verify that 0.52 is the value of emergent
for _ in range(100):
u.spec.update_avg_l(u)
self.assertTrue(np.allclose(1.64, u.avg_l, rtol=0.1, atol=0.1))
def test_forced_act(self):
"""Test that forcing activity behaves as expected"""
u = leabra.Unit()
u.force_activity(0.5)
u.calculate_net_in()
u.cycle()
self.assertEqual(u.act, 0.5)
# is it maintained without call to add_excitatory() ?
for _ in range(10):
u.calculate_net_in()
u.cycle()
self.assertEqual(u.act, 0.5)
u.add_excitatory(0.5)
u.calculate_net_in()
u.cycle()
self.assertFalse(u.act == 0.5)
def test_emergent_xx1(self):
"""Test quantitative equivalence with emergent on the xx1 function."""
xx1_data = data.parse_xy('convolve_nxx1.txt')
spec = leabra.UnitSpec(adapt_on=False, noisy_act=True)
receiver = leabra.Unit(spec=spec)
receiver.cycle() # create the noisy_xx1 convolution
xs, ys = spec._nxx1_conv
check = True
for k, (pys, ems) in enumerate([(xs, xx1_data['x']),
(ys, xx1_data['y'])]):
for t, (py, em) in enumerate(zip(pys, ems)):
if not np.allclose(py, em, rtol=1e-05, atol=1e-08):
print('{}:{} [py] {:.10f} != {:.10f} [emergent]'.format(
'x' if k == 0 else 'y', t, py, em))
check = False
self.assertTrue(check)