def test_modell():
timesteps = 12000
neurons_input = 180
neurons_output = 180
neurons = neurons_input + neurons_output
spiking_model = spiking.SRM(neurons=neurons, threshold=1.0, t_current=0.3,
t_membrane=20, eta_reset=5, verbose=False)
learning_model = learning.STDP(eta=0.05, w_in=0.0, w_out=-0.05, tau=10.0, window_size=5, verbose=False)
weights = np.zeros((neurons, neurons))
# Input Neurons are fully connected to the output neurons
mu, sigma = 0.5, 0.05
weights[:neurons_input, neurons_input:] = sigma * np.random.randn(neurons_input, neurons_output) + mu
# Output Neurons are fully connected to each other with inhibitory weights
mu, sigma = -0.5, 0.05
weights_output = sigma * np.random.randn(neurons_output, neurons_output) + mu
weights[neurons_input:, neurons_input:] = weights_output
# Output neurons have exhibitory weights for their neighboring neurons
mu, sigma = 0.5, 0.05
b = np.eye(neurons_output) * (sigma * np.random.randn(neurons_output, 1) + mu)
weights[neurons_input:, neurons_input:] += np.roll(b, 1, 1)
weights_output = weights[neurons_input:, neurons_input:].copy()
spikes = np.zeros((neurons, timesteps), dtype=bool)
# Generate some spikes in the input layer
for i in range(neurons_input):
spikes[i] = tools.sound(timesteps, 20 * i, 0.4, 50)
spikes[i + neurons_input, 20 * i - 5:20 * i + 5] = True
for i in range(neurons_input - 1, -1, -1):
spikes[i] = spikes[i] + tools.sound(timesteps, neurons_input * 20 + 20 * i, 0.4, 50)
spikes[i + neurons_input, neurons_input * 20 + 20 * i - 5: neurons_input * 20 + 20 * i + 5] = True
# Weight plot
weightplotter = plotting.WeightHeatmapAnimation()
weightplotter.add(weights)
save_interval = 10
for t in range(timesteps):
print("Timestep: ", t)
if weightplotter and t % save_interval == 0:
weightplotter.add(weights)
spiking_model.simulate(spikes, weights, t)
learning_model.weight_change(spikes, weights, t)
# Reset output layer weights
weights[neurons_input:, neurons_input:] = weights_output
weightplotter.show_animation()
def test_jeffress():
neurons = 11
timesteps = 100
ax_delays = np.array([0, 5, 15, 25, 0, 25, 15, 5, 0, 0, 0])
model = spiking.SRM_X(neurons=neurons, threshold=np.array([1]*neurons), t_current=np.array([5]*neurons),
t_membrane=np.array([10]*neurons), eta_reset=np.array([2.0]*neurons), ax_delay=ax_delays)
weights = np.zeros((neurons, neurons))
# Connect input layer
weights[0, (1, 2, 3)] = 1
weights[4, (5, 6, 7)] = 1
# Connect to output layer
weights[(1, 5), 8] = 1.05
weights[(2, 6), 9] = 1.05
weights[(3, 7), 10] = 1.05
print(weights)
spiketrain = np.zeros((neurons, timesteps), dtype=bool)
manual_spiketrain = True
if manual_spiketrain:
spiketrain[0, (20, 25, 30)] = 1
spiketrain[4, (0, 5, 10)] = 1
else:
spiketrain[0,:] = tools.sound(timesteps, 85, 0.6, 4)
spiketrain[4,:] = tools.sound(timesteps, 60, 0.6, 4)
psth = plotting.PSTH(spiketrain, binsize=5)
curr = plotting.CurrentPlot(4)
for t in range(timesteps):
current = model.check_spikes(spiketrain, weights, t)
curr.add(current[[0, 3, 7, 10]])
psth.show_plot()
curr.show_plot()
def test_modell():
timesteps = 12000
neurons_input = 180
neurons_output = 180
neurons = neurons_input + neurons_output
spiking_model = spiking.SRM(neurons=neurons,
threshold=1.0,
t_current=0.3,
t_membrane=20,
eta_reset=5,
verbose=False)
learning_model = learning.STDP(eta=0.05,
w_in=0.0,
w_out=-0.05,
tau=10.0,
window_size=5,
verbose=False)
weights = np.zeros((neurons, neurons))
# Input Neurons are fully connected to the output neurons
mu, sigma = 0.5, 0.05
weights[:neurons_input, neurons_input:] = sigma * np.random.randn(
neurons_input, neurons_output) + mu
# Output Neurons are fully connected to each other with inhibitory weights
mu, sigma = -0.5, 0.05
weights_output = sigma * np.random.randn(neurons_output,
neurons_output) + mu
weights[neurons_input:, neurons_input:] = weights_output
# Output neurons have exhibitory weights for their neighboring neurons
mu, sigma = 0.5, 0.05
b = np.eye(neurons_output) * (sigma * np.random.randn(neurons_output, 1) +
mu)
weights[neurons_input:, neurons_input:] += np.roll(b, 1, 1)
weights_output = weights[neurons_input:, neurons_input:].copy()
spikes = np.zeros((neurons, timesteps), dtype=bool)
# Generate some spikes in the input layer
for i in range(neurons_input):
spikes[i] = tools.sound(timesteps, 20 * i, 0.4, 50)
spikes[i + neurons_input, 20 * i - 5:20 * i + 5] = True
for i in range(neurons_input - 1, -1, -1):
spikes[i] = spikes[i] + tools.sound(
timesteps, neurons_input * 20 + 20 * i, 0.4, 50)
spikes[i + neurons_input, neurons_input * 20 + 20 * i -
5:neurons_input * 20 + 20 * i + 5] = True
# Weight plot
weightplotter = plotting.WeightHeatmapAnimation()
weightplotter.add(weights)
save_interval = 10
for t in range(timesteps):
print("Timestep: ", t)
if weightplotter and t % save_interval == 0:
weightplotter.add(weights)
spiking_model.simulate(spikes, weights, t)
learning_model.weight_change(spikes, weights, t)
# Reset output layer weights
weights[neurons_input:, neurons_input:] = weights_output
weightplotter.show_animation()