def link_field_to_neuron(entry_fields: dict, name: str, neuron: neurons.Neuron,
notify_neuron: Callable):
"""Link the fields associated with a button to its neuron.
Args:
entry_fields: A mapping of field names to instances of EntryField.
name: …of the field being mapped to the neuron.
neuron:
notify_neuron:
"""
entry_fields[name].textvariable.trace_add('write', notify_neuron)
neuron.register_event(name)
def calibrate_current(self, ttfs):
for current in range(0, 10000, 1):
neuron = Neuron()
for time in range(0, ttfs + 1):
neuron.time_step(current / 10., time)
if neuron.has_spiked(): break
if neuron.last_spike == ttfs:
return current / 10.
def neuron_linker(self, internal_name: str, neuron: neurons.Neuron,
neuron_callback: Callable, initial_state: bool = False):
"""Set a neuron callback which will be called whenever the field is changed by the user.
Args:
internal_name: Name of widget. The neuron will be notified whenever this widget is
changed by the user.
neuron:
neuron_callback: This will be set as the trace_add method of the field's textvariable.
initial_state:
Returns:
"""
self.entry_fields[internal_name].textvariable.trace_add('write',
neuron_callback(internal_name, neuron))
neuron.register_event(internal_name, initial_state)
def sum_epsps(self, time, neuron):
current = 0
prev = neuron.i_layer-1
for _pre in self.layers[prev]:
if _pre.has_spiked():
ls, wt = _pre.last_spike, self.synapses[prev][_pre][neuron].weight
if neuron.i_layer == 2 and _pre.type == 'inhibitory': wt *= -1
current += wt * Neuron().v_max * math.e**(-(time-ls)/self.time_const)
return current
def __init__(self, n_layers):
self.time_const = 3.
self.learn_rate = 7.
self.excit_ratio = 10
self.t_window = 50
self.trained = defaultdict()
self.layers = [[Neuron() for i in range(n)] for n in n_layers]
self.synapses = [
np.random.rand(len1, len2) for len1, len2 in pairwise(n_layers)
]
def calibrate_current(self, rate):
""" Rate must be less than self.t_window """
for current in range(0, 1000, 1):
neuron = Neuron()
spikes = 0
for c_time in range(1, self.t_window + 1):
if neuron.time_step(current / 10., c_time) == 1:
spikes += 1
if spikes == rate: return current / 10.
if spikes == self.t_window: break
return -1
def __init__(self, n_inputs):
s = self
s.lambduh = .25 # learning rate
s.on_ttfs = 4 # ms when 'on' neurons should spike
s.t_threshold = 25 # time to wait for output to spike
s.V_rest, s.V_th = 0, n_inputs
s.neurons = [Neuron(i_neuron=i) for i in range(n_inputs)]
s.synapses = [random.uniform(-1, 1) for neuron in s.neurons]
while len(filter(lambda w: w < 0, s.synapses)) >= math.sqrt(n_inputs):
s.synapses = [random.uniform(-1, 1) for neuron in s.neurons] # limit inhibs
def __init__(self, n_inputs):
s = self
s.lambduh = .25 # learning rate
s.on_ttfs = 3 # ms when 'on' neurons should spike
s.t_threshold = 10 # time to wait for output to spike
s.V_rest, s.V_th = 0, n_inputs
s.neurons = [Neuron(i_neuron=i) for i in range(n_inputs)]
s.synapses = [random.uniform(-1, 1) for neuron in s.neurons]
for i in range(len(s.neurons)):
if i >= math.sqrt(n_inputs):
s.synapses[i] = random.uniform(0, 1) # limit inhibs
def sum_epsps(self, c_time, i_c_layer, i_c_neuron):
current = 0
c_layer = self.layers[i_c_layer]
p_layer = self.layers[i_c_layer - 1]
c_synapse = self.synapses[i_c_layer - 1]
for i_p_neuron in range(len(p_layer)):
p_neuron = p_layer[i_p_neuron]
p_spike = p_neuron.last_spike
if p_spike != -1: # ttfs exists
weight = c_synapse[i_p_neuron][i_c_neuron]
change = weight * Neuron().v_max * math.e**(
-(c_time - p_spike) / self.time_const)
#if VERBOSE: print('\tadding epsp to current: {}'.format(change))
current += change
return current
def __init__(self, n_layers):
s = self
s.time_const = 3.
s.learn_rate = 7.
s.excit_ratio = 2
s.trained = defaultdict()
s.layers = [[Neuron(i_layer=i_layer, i_neuron=i_neuron)
for i_neuron in range(n_layers[i_layer])]
for i_layer in range(len(n_layers))]
s.synapses = [s.gen_synapse_layer(pre_layer, post_layer)
for pre_layer, post_layer in pairwise(s.layers)]
# make some of hidden layer inhibitory
#hid_out = s.synapses[len(s.layers)/2]
for layer in s.layers[0:int(math.ceil(len(s.layers)/s.excit_ratio))]:
for neuron in layer: neuron.type = 'inhibitory'
def __init__(self, num_neurons, bias):
self.bias = bias
self.neurons = []
for _ in range(num_neurons):
self.neurons.append(Neuron(self.bias))
from neurons import Neuron, InputNeuron, OutputNeuron
i1 = InputNeuron(0)
i2 = InputNeuron(1)
i3 = InputNeuron(0)
# learning rate, thresholds of connected neurons, connected neurons, initial connection weights
h1 = Neuron(0.1, 0.5, [i1], [1])
h2 = Neuron(0.1, 0.3, [i2], [1])
h3 = Neuron(0.1, 0.8, [i3], [1])
h4 = Neuron(0.1, 0.6, [h1, h2, h3], [0.3, -0.4, 0.7])
o = OutputNeuron(0.1, [h4], [0.3])
print(o.fire())
for _ in range(100):
h4.calc_error_delta_rule(1, 1)
print(o.fire())
from neurons import Neuron, Network, Synapse, InhibitorySynapse, ActivityPattern n = Neuron() print(n.state.dendrites) n.update() print(n.state.axon) n.excite() print(n.state.axon) n.inhibit() print(n.state.axon) N = Network() N.add_neuron() N.add_neuron() print(N.neurons[0].state.axon) print(N.neurons[1].state.axon) N = Network() n1 = N.add_neuron() print(N.history) N.ntrace() print(N.history) n1.excite() print(N.history)
def generate(self, inputs):
neurons = []
for i in [1] * self.neurons:
neurons.append(Neuron(inputs, self.activation))
self.neurons = neurons
