def test_predict(self):
target = Neuron(tf.constant([1, 2, 3]), tf.constant(4))
prediction = target.predict(
tf.constant([[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12]]))
tf.debugging.assert_equal(prediction, tf.constant([18, 36, 54, 72]))
class OurNeuralNetwork:
'''
A neural network with:
- 2 inputs
- a hidden layer with 2 neurons (h1, h2)
- an output layer with 1 neuron (o1)
Each neuron has the same weights and bias:
- w = [0, 1]
- b = 0
'''
def __init__(self):
weights = np.array([0, 1])
bias = 0
# The Neuron class here is from the previous section
self.h1 = Neuron(weights, bias)
self.h2 = Neuron(weights, bias)
self.o1 = Neuron(weights, bias)
def feedforward(self, x):
out_h1 = self.h1.feedforward(x)
out_h2 = self.h2.feedforward(x)
# The inputs for o1 are the outputs from h1 and h2
out_o1 = self.o1.feedforward(np.array([out_h1, out_h2]))
return out_o1
def test_integrator(self):
neuron = Neuron(
weights=[1, 2, 3],
transfer_function=BaseTransferFunction, # does nothing
)
self.assertEqual(neuron.integrator([1, 1, 1]), 6)
def __init__(self, input_size, lrn_rate=1, activation=signal):
"""'input_size' is the length of the input.
'lrn_rate' is the learning rate.
"""
self.neuron = Neuron([0] * input_size, 0, activation)
self.lrn_rate = lrn_rate
self.fire = self.neuron.fire
class Perceptron(object):
def __init__(self, input_size, lrn_rate=1):
"""'input_size' is the length of the input.
'lrn_rate' is the learning rate.
"""
self.neuron = Neuron([0]*input_size, 0, signal)
self.lrn_rate = lrn_rate
self.fire = self.neuron.fire
def training(self, examples):
epochs = 0
while True:
epochs = epochs + 1
error_count = 0
for (input_vector, desired_output) in examples:
actual_output = self.neuron.fire(input_vector)
error = desired_output - actual_output
if error != 0:
learned = self.lrn_rate*error
self.neuron.update(input_vector, learned)
error_count = error_count + 1
if error_count == 0:
break
return epochs
def __str__(self):
ret = 'lrn_rate: %s' % self.lrn_rate
ret = '%s\n%s' % (ret, self.neuron.__str__())
return ret
def __init__(self,
map_dimensions,
data_dimension,
kernel_func,
nb_clusters=0,
verbose=False):
self._verbose = verbose
self._map_dimensions = []
self._map_dimensions[:] = map_dimensions[:]
self._data_dimension = data_dimension
self._dataset = None
self._kernel = kernel_func
self._nb_clusters = nb_clusters
self._neurons = []
self._clusters = []
for position in generate_positions(map_dimensions):
# Initialize _neurons positions in map
neuron = Neuron(position=position)
# Initialize _neurons coordinates
coords = []
for i in xrange(data_dimension):
coords.append(random.random() * 10)
neuron.set_data(coords)
self._neurons.append(neuron)
def test_step_false(self):
neuron = Neuron(
weights=[1, 2, 3],
transfer_function=StepTransferFunction,
function=lambda p: p >= 7, # any function can be used here
)
self.assertEqual(neuron.run([1, 1, 1]), 0)
def testSinglePreviousEvaluate(self): previousNeuron = InputNeuron() previousNeuron.setValue(1) previousRow = [previousNeuron] neuron = Neuron(previousRow) self.assertGreater(neuron.evaluate(), 1/2)
def __init__(self,
layerName,
inputArr,
neuronCount=0,
prevNeuronCount=0,
activation="relu",
mode="init_rand"):
self.layerName = layerName
self.inputArr = inputArr
self.activation = activation
self.mode = mode
self.neurons = []
if mode == "init_rand":
self.neuronCount = neuronCount
self.prevNeuronCount = prevNeuronCount
for i in range(self.neuronCount):
neuron = Neuron(input=self.inputArr,
weight=np.random.rand(self.prevNeuronCount),
bias=np.random.randint(-5, 5))
self.neurons.append(neuron)
elif mode == "init_read":
file = open(self.layerName + ".txt", "r")
data = json.load(file)
self.neuronCount = data["neuronCount"]
self.prevNeuronCount = data["prevNeuronCount"]
for i in range(self.neuronCount):
weightArr = data["weight" + str(i + 1)]
biasVal = data["bias" + str(i + 1)]
neuron = Neuron(self.inputArr, weightArr, biasVal)
self.neurons.append(neuron)
def play(self, hand, opponent_card_weigth, who_made_first):
self.who_made_first = who_made_first
cards_in_hand_weights = [k[2] for k in self.cards_in_hand]
cards_in_hand_mean = sum(cards_in_hand_weights) / len(
cards_in_hand_weights)
neuron_entries = [[
who_made_first, opponent_card_weigth, cards_in_hand_mean, hand
]]
neuron_value = Neuron(neuron_entries).exec()
have_major_card = False
for card in self.cards_in_hand:
if card[2] >= opponent_card_weigth:
have_major_card = True
if not have_major_card:
return cards_in_hand_weights.index(min(cards_in_hand_weights))
if round(neuron_value) == 1:
if self.can_trucar:
return self.trucar()
k = 0
for card_in_hand in self.cards_in_hand:
entries_choice = [[
who_made_first,
self.cards_weight(self.opponent_cards_played), card_in_hand[2],
hand
]]
choice_neuron = Neuron(entries_choice).exec()
if round(choice_neuron) == 1:
return k
k += 1
return 0
def build_layer(weights_filename, biases_filename, prev_layer_size):
new_layer = []
# Build neurons with starting weights
with open(weights_filename, 'r') as weights_file:
reader = csv.reader(weights_file)
# Iterate CSV rows / neurons
for row in reader:
new_neuron = Neuron()
weights = []
# Iterate weights
for i in range(prev_layer_size):
weights.append(float(row[i]))
new_neuron.set_weights(weights)
new_layer.append(new_neuron)
# Add starting biases
with open(biases_filename, 'r') as biases_file:
reader = csv.reader(biases_file)
# Iterate CSV rows / neurons
for index, row in enumerate(reader):
for bias in row:
new_layer[index].set_bias(float(bias))
return new_layer
def half_adder_train(inputs, user_iteration, learning_rate):
targets = [[0, 0], [0, 1], [0, 1], [1, 0]]
half_adder = NeuronNetwork([
NeuronLayer([
Neuron("1", [randint(-1, 1), randint(-1, 1)], randint(-1, 1)),
Neuron("2", [randint(-1, 1), randint(-1, 1)], randint(-1, 1))
]),
NeuronLayer([
Neuron("3", [randint(-1, 1), randint(-1, 1)], randint(-1, 1)),
Neuron("4", [randint(-1, 1), randint(-1, 1)], randint(-1, 1))
])
])
iterations, errors, outputs = half_adder.train(inputs, targets,
learning_rate,
user_iteration)
print(
f"============ | Half adder | ============\n"
f"After {iterations} iterations:\n"
f"Errors: {errors}\n"
f"Inputs: {inputs}\n"
f"Outputs: {outputs}\n"
f"Targets: {targets}\n"
f"Weights:",
end=' ')
layer = half_adder.layers[-1]
print([neuron.weights for neuron in layer.neurons])
print(f"Bias: {[neuron.bias for neuron in layer.neurons]}\n\n")
class Perceptron(object):
def __init__(self, input_size, lrn_rate=1):
"""'input_size' is the length of the input.
'lrn_rate' is the learning rate.
"""
self.neuron = Neuron([0] * input_size, 0, signal)
self.lrn_rate = lrn_rate
self.fire = self.neuron.fire
def training(self, examples):
epochs = 0
while True:
epochs = epochs + 1
error_count = 0
for (input_vector, desired_output) in examples:
actual_output = self.neuron.fire(input_vector)
error = desired_output - actual_output
if error != 0:
learned = self.lrn_rate * error
self.neuron.update(input_vector, learned)
error_count = error_count + 1
if error_count == 0:
break
return epochs
def __str__(self):
ret = 'lrn_rate: %s' % self.lrn_rate
ret = '%s\n%s' % (ret, self.neuron.__str__())
return ret
class Merge(threading.Thread):
URI = ""
eqc = 0 # counter
def __init__(self, __uri):
threading.Thread.__init__(self)
self.URI = __uri
self.n = Neuron()
self.n.start()
self.eqc = 0
def run(self):
while True:
for entry in measEntries:
self.n.add(entry.Weight)
measEntries.clear()
if self.n.FireState == True:
self.eqc = self.eqc + 1
if self.eqc > 20:
self.eqc = 20
else:
if self.eqc > 0:
self.eqc = self.eqc - 1
if self.eqc > 10:
print("EARTHQUAKE!!!")
requests.get(self.URI + "/warning?description=Earthquake")
elif self.eqc < 5:
requests.get(self.URI + "/warning?description=ok")
time.sleep(0.1)
def test_set_error_output_layer(self):
neuron = Neuron(0, 0, [0.05, 0.05], [1, 1])
neuron.output = 0.518979
neuron.is_output_layer = True
neuron.set_output_layer_error(0)
self.assertEquals(-0.12955, round_to(neuron.delta_val, 5))
def add_layer(self, neurons, layer_number):
"""
:param neurons:
:param layer_number:
:return:
"""
bias_value = random.randint(1,10)
bias = Neuron(activation_func=lambda x: 0, activation_prime=lambda x: 0, isBias=True)
bias.y_output = bias_value
neurons.append(bias)
self.layers[layer_number] = neurons
if layer_number == 0:
return
if layer_number > self.max_layer:
self.max_layer = layer_number
for input in self.layers[layer_number - 1]:
for output in neurons:
output.add_input_reference(input)
if not output.isBias:
weight = self.randomize_weight()
input.add_output_connection(output, weight)
def __init__(self, joysize, painsize):
defaultweight = 0.3
preneurons = []
#create neurons
for i in range(joysize):
tempneuron = Neuron('amygdala' + str(i), 'joyneurons',
actthreshold)
preneurons.append(tempneuron)
self.joyneurons = preneurons
preneurons = []
for i in range(painsize):
tempneuron = Neuron('amygdala' + str(i + joysize), 'painneurons',
actthreshold)
preneurons.append(tempneuron)
self.painneurons = preneurons
preneurons = []
for neuron in self.joyneurons:
preneurons.append(neuron)
for neuron in self.painneurons:
preneurons.append(neuron)
self.neurons = preneurons
self.size = joysize + painsize
self.joysize = joysize
self.painsize = painsize
#connect neurons
for neuron1 in self.neurons:
for neuron2 in self.neurons:
if neuron1 is not neuron2:
neuron1.add_inconnects([neuron2, defaultweight])
neuron2.add_outconnects([neuron1, defaultweight])
def reset(self):
"""Resets the state of the neuron
"""
Neuron.reset(self)
self.voltage.set_value(numpy.zeros(self.size).astype('float32'))
self.refractory_time.set_value(
numpy.zeros(self.size).astype('float32'))
def test_update_weights(self):
neuron = Neuron(0, 0, [0.05, 0.05], [0.519053, 1])
neuron.delta_val = -0.1295578
neuron.update_weights(0.001)
self.assertEquals(0.0499327526, round(neuron.weights[0], 10))
self.assertEquals(0.0498704, round(neuron.weights[1], 7))
def Adder():
A = 0
B = 0
weights = -0.5
thresholds = -0.5
gate1 = Neuron([[A, weights], [B, weights]], threshold=thresholds).calc()
gate2 = Neuron([[A, weights], [gate1, weights]],
threshold=thresholds).calc()
gate3 = Neuron([[gate1, weights], [B, weights]],
threshold=thresholds).calc()
gate4 = Neuron([[gate2, weights], [gate3, weights]],
threshold=thresholds).calc()
gate5 = Neuron([[gate1, weights], [gate1, weights]],
threshold=thresholds).calc()
carry = gate5
output = gate4
print("Carry : " + str(carry))
print("Output: " + str(output))
def __init__(self,
nodes,
initial_weights=None,
bias=None,
neuron_type="linear"):
self._ntype = neuron_type
self._neurons = Neuron(neuron_type)
assert (len(nodes) == 2), "param nodes is a tuple/list of length 2"
assert (type(nodes[0]) == int)
assert (type(nodes[1]) == int)
self._nodes = nodes
self._init_weights = initial_weights
self._bias = bias
grad_fn = dict()
grad_fn = {
"linear": self._linear_layer_grad,
"logistic": self._logistic_layer_grad
}
self.grad = grad_fn[neuron_type]
if self._init_weights is None:
self._init_weights = np.random.rand(nodes[0], nodes[1])
if self._bias is None:
self._bias = np.ones(self._nodes[1])
assert (len(self._init_weights.T) == self._nodes[1])
assert (len(self._bias) == self._nodes[1])
self._state = np.zeros(self._nodes[1])
self._bias_weights = np.ones(self._nodes[1])
self._weights = self._init_weights
def test_step_true(self):
neuron = Neuron(
weights=[1, 2, 3],
transfer_function=StepTransferFunction,
)
self.assertEqual(neuron.run([1, 2, 3]), 1)
def xor_port_train(inputs, user_iteration, learning_rate):
targets = [0, 1, 1, 0]
xor_port = NeuronNetwork([
NeuronLayer([
Neuron("Nor gate", [randint(-1, 1), randint(-1, 1)],
randint(-1, 1)),
Neuron("And gate", [randint(-1, 1), randint(-1, 1)],
randint(-1, 1))
]),
NeuronLayer([
Neuron("Nor gate", [randint(-1, 1), randint(-1, 1)],
randint(-1, 1))
])
])
iterations, errors, outputs = xor_port.train(inputs, targets,
learning_rate, user_iteration)
print(
f"============ | Xor gate | ============\n"
f"After {iterations} iterations:\n"
f"Errors: {errors}\n"
f"Inputs: {inputs}\n"
f"Outputs: {outputs}\n"
f"Targets: {targets}\n"
f"Weights:",
end=' ')
layer = xor_port.layers[-1]
print([neuron.weights for neuron in layer.neurons])
print(f"Bias: {[neuron.bias for neuron in layer.neurons]}\n\n")
def test_updateBias(self):
bias1 = tf.constant(7)
target = Neuron([4, 5, 6], bias1)
self.assertEqual(target._bias, bias1)
bias2 = tf.constant(8)
target.updateBias(bias2)
self.assertEqual(target._bias, bias2)
def loadmat(self):
"""Load neurons from a single .mat file"""
self.header = MATHeader()
nrecs = self.header.read(self.path)
for nrec in nrecs:
neuron = Neuron(self.path, sort=self)
neuron.loadmat(nrec)
self.alln[neuron.id] = neuron # save it
def __init__(self, input_size, size, weights=None):
if weights is None:
self.neurons = [
Neuron(np.random.rand(input_size + 1) - 0.5)
for _ in range(size)
]
else:
self.neurons = [Neuron(weights[i]) for i in range(size)]
def setUp(self):
# List of Neuron objects
neur_1 = Neuron()
neur_2 = Neuron()
self.neurons = [neur_1, neur_2]
# Weight matrix
# Input goes 1, 1 goes to 2, 2 goes to output
self.weights = np.array([[1, 0, 0], [0, 1, 0]])
def __init__(self, size, dt=0.001, t_rc=0.02, t_ref=0.002):
Neuron.__init__(self, size, dt)
self.t_rc = t_rc
self.t_ref = t_ref
self.voltage = theano.shared(
numpy.zeros(size).astype('float32')) # internal variables
self.refractory_time = theano.shared(
numpy.zeros(size).astype('float32')) # internal variables
def create_layer(self, n_num, last_layer=False):
self.layers.append([
Neuron(None if len(self.layers) == 0 else self.layers[-1])
for i in range(0, n_num)
])
# cria um neuronio de bias
if not last_layer:
self.layers[-1].append(Neuron(None, bias_neuron=True))
def test_sigmoid(self):
neuron = Neuron(
weights=[1, 2, 3],
transfer_function=SigmoidTransferFunction,
)
v = neuron.run([0, 0, 0])
self.assertEqual(v, 0.5)
def load(path):
"""
Loads a neural network from a json file
@param (String) path - The path to load the neural network from
@returns (Network) - The neural network that was loaded
"""
network = Network()
try:
with open(path, "r+") as f:
network_data = "\n".join(f.readlines())
network_json = json.loads(network_data)
layers = network_json["layers"]
# For every layer in the network ...
for layer in layers:
neurons = []
# For every neuron in the layer ...
for neuron in layer["neurons"]:
weights = neuron["weights"]
bias = neuron["bias"]
activation = neuron["activation"]
# Choose the proper activation function and corresponding derivative
activation_func = None
derivative_func = None
if activation == Network.LINEAR:
activation_func = Network.ACTIVATION_LINEAR
derivative_func = Network.DERIVATIVE_LINEAR
elif activation == Network.SIGMOID:
activation_func = Network.ACTIVATION_SIGMOID
derivative_func = Network.DERIVATIVE_SIGMOID
elif activation == Network.TANH:
activation_func = Network.ACTIVATION_TANH
derivative_func = Network.DERIVATIVE_TANH
elif activation == Network.STEP:
activation_func = Network.ACTIVATION_STEP
derivative_func = Network.DERIVATIVE_STEP
# Create a neuron with the desired info
neuron = Neuron(0, activation_func, derivative_func)
neuron.weights = weights
neuron.bias = bias
# Add the processed neuron to the collection
neurons.append(neuron)
# Create a layer with the desired neurons
layer = Layer(0, 0, None, None)
layer.neurons = neurons
# Add the processed layer to the collection
network.layers.append(layer)
except:
raise Exception("Invalid Neural Network File @ {}!".format(path))
return network
def load(path):
"""
Loads a neural network from a json file
@param (String) path - The path to load the neural network from
@returns (Network) - The neural network that was loaded
"""
network = Network()
try:
with open(path, "r+") as f:
network_data = "\n".join(f.readlines())
network_json = json.loads(network_data)
layers = network_json["layers"]
# For every layer in the network ...
for layer in layers:
neurons = []
# For every neuron in the layer ...
for neuron in layer["neurons"]:
weights = neuron["weights"]
bias = neuron["bias"]
activation = neuron["activation"]
# Choose the proper activation function and corresponding derivative
activation_func = None
derivative_func = None
if activation == Network.LINEAR:
activation_func = Network.ACTIVATION_LINEAR
derivative_func = Network.DERIVATIVE_LINEAR
elif activation == Network.SIGMOID:
activation_func = Network.ACTIVATION_SIGMOID
derivative_func = Network.DERIVATIVE_SIGMOID
elif activation == Network.TANH:
activation_func = Network.ACTIVATION_TANH
derivative_func = Network.DERIVATIVE_TANH
elif activation == Network.STEP:
activation_func = Network.ACTIVATION_STEP
derivative_func = Network.DERIVATIVE_STEP
# Create a neuron with the desired info
neuron = Neuron(0, activation_func, derivative_func)
neuron.weights = weights
neuron.bias = bias
# Add the processed neuron to the collection
neurons.append(neuron)
# Create a layer with the desired neurons
layer = Layer(0, 0, None, None)
layer.neurons = neurons
# Add the processed layer to the collection
network.layers.append(layer)
except:
raise Exception("Invalid Neural Network File @ {}!".format(path))
return network
def test_updateWeights(self):
weights1 = tf.constant([4, 5, 6])
target = Neuron(weights1, 7)
tf.debugging.assert_equal(target._weights, weights1)
weights2 = tf.constant([8, 9, 10])
target.updateWeights(weights2)
tf.debugging.assert_equal(target._weights, weights2)
def __init__(self, num_inputs, num_hidden_layers, num_neurons):
self.fitness = None
self.output = Neuron()
self.inputs = [Neuron() for _ in range(num_inputs)]
self.hidden_layers = [[Neuron() for _ in range(num_neurons)]
for _ in range(num_hidden_layers)]
self.synapses = [[] for _ in range(num_hidden_layers + 1)]
self.num_hidden_layers = num_hidden_layers
self.init_synapses()
def __init__(self, size, *pargs):
self.base = []
if pargs:
layer, rand = pargs
for i in range(size):
self.base.append(Neuron(layer, rand))
else:
for i in range(size):
self.base.append(Neuron())
def __init__(self, inputs, weights, bias):
Neuron.__init__(self, inputs)
# NOTE: The weights and bias properties here are not
# numbers, but rather references to other neurons.
# The weight and bias values are stored within the
# respective neurons.
self.weights = weights
self.bias = bias
def test_neuron():
print('Testing a neuron...')
activations = ActivationFunctions()
weights = np.array([3, 1]) # w1 = 3, w2 = 1
bias = -1 # b = -1
n = Neuron(weights, bias, activations.sigmoid)
x = np.array([4, 1]) # x1 = 4, x2 = 1
print(n.show_configuration())
print('Input: {} --> Result: {}'.format(x, n.feedforward(x))) # 0.9999938558253978
def generation_crossover(self, datas):
new_gen = []
for i in range(0, self.count_individuals):
should_mutate = random() < self.mutation_prob
if should_mutate:
self.root_neurons[i].mutate()
should_struct_mutate = random() < self.mutation_struct_prob
if should_struct_mutate:
self.root_neurons[i].mutate_struct()
sorted(datas, key=lambda d: d.score, reverse=True)
for i in range(0, len(datas)):
for j in range(0, len(datas)):
if i == j:
continue
root_synapses = self.root_neurons[i].synapses
for syn in root_synapses:
syn.weight = (
syn.weight + self.root_neurons[j].synapses[root_synapses.index(syn)].weight) / 2
root_neuron = Neuron(0, root_synapses)
f_genes, s_genes = []
self.root_neurons[i].iterate_children_recursive(
lambda neuron, next_neuron, syn: f_genes.append((neuron, next_neuron, syn)))
self.root_neurons[j].iterate_children_recursive(
lambda neuron, next_neuron, syn: s_genes.append((neuron, next_neuron, syn)))
for gene in f_genes:
for o_gene in s_genes:
if gene[0].id == o_gene[0].id and gene[1].id == o_gene[1].id or random() < self.crossover_unique_gene_transfer_prob:
match_neuron = root_neuron.find_child(
lambda n: n.id == o_gene[0].id) if o_gene[0].id != 0 else False
ancestor_genes = root_neuron
new_syn = Synapse(o_gene[2].label
(gene[2].input + o_gene[2].input) / 2, (gene[2].weight + o_gene[2].weight) / 2, [])
if match_neuron is None:
match_neuron = Neuron(o_gene[0].id, [new_syn])
match_ancestor_genes = filter(
lambda g: g[1].id == match_neuron.id, s_genes)
if len(match_ancestor_genes) > 0:
ancestor_genes = match_ancestor_genes[0]
new_neuron = Neuron(o_gene[1].id + len(map(lambda n:
map(lambda s: s.next_neurons, n.synapses), match_neuron)) + 1, [new_syn])
if ancestor_genes is root_neuron:
root_neuron.synapses.append(
Synapse(new_syn.label, new_syn.input, new_syn.weight, [new_neuron]))
else:
ancestor_genes[2].next_neurons.append(
new_neuron)
if len(new_gen) == self.count_individuals:
break
self.root_neurons = new_gen
self.adjust_weights(datas)
def loadptcs(self):
"""Load neurons from a single .ptcs file"""
self.header = PTCSHeader()
with open(self.path, 'rb') as f:
self.header.read(f)
for i in range(self.header.nneurons):
neuron = Neuron(self.path, sort=self)
neuron.loadptcs(f, self.header)
self.alln[neuron.id] = neuron # save it
assert eof(f), 'File %s has unexpected length' % self.path
def init(self, neurons_num=3, inputs=3, activation_function="sigmoid"):
self.inputs_num = inputs
self.neurons_num = neurons_num
self.activation_function = activation_function
self.neurons = []
for i in range(self.neurons_num):
neuron = Neuron()
neuron.init(self.inputs_num, self.activation_function)
self.neurons.append(neuron)
def __init__(self):
self.rs = RS_232(wrk_dir='/tmp/Cortix', filename='ir_7040')
self.neuron = Neuron(task=self.rs.ir_7040,wrk_dir='/tmp/Cortix',name='ir_7040')\
self.socket_list = []
self.init_port = 60000
self.port = 60001
#self.host= str(socket.gethostbyname(socket.gethostname()))
self.host = '10.253.90.99'
print(self.host)
def test_3(steps):
weights = 1
print "Target converse {0}, {1} steps".format(weights, steps)
neuron = Neuron(weights, sigm, sigmp, error)
errors = []
for i in range(steps):
inputs = [random.random() for r in range(weights)]
target = 1.0 - inputs[0]
neuron.learn_1(inputs, target)
errors.append(neuron.last_error)
print report(errors)
def test_feed_forward(self):
neuron = Neuron(0, 0, [0.05, 0.05], [1, 1])
next_node1 = Neuron(0, 0, [0.05, 0.05], [0, 0])
next_node2 = Neuron(0, 0, [0.05, 0.05], [0, 0])
nodes = [next_node1, next_node2]
neuron.feed_forward(nodes)
self.assertEquals(0.524, round_to(nodes[0].inputs[0], 3))
self.assertEquals(0, round_to(nodes[0].inputs[1], 3))
self.assertEquals(0.524, round_to(nodes[1].inputs[0], 3))
self.assertEquals(0, round_to(nodes[1].inputs[1], 3))
def __init__(self, size, dt=0.001, tau_rc=0.02, tau_ref=0.002):
"""Constructor for a set of LIF rate neuron
:param int size: number of neurons in set
:param float dt: timestep for neuron update function
:param float t_rc: the RC time constant
:param float tau_ref: refractory period length (s)
"""
Neuron.__init__(self, size, dt)
self.tau_rc = tau_rc
self.tau_ref = tau_ref
def test_1(steps):
weights = 3
print "Linear combination of weights {0}, {1} steps".format(weights, steps)
neuron = Neuron(weights, sigm, sigmp, error)
errors = []
for i in range(steps):
inputs = [random.random() for r in range(weights)]
target = 2*inputs[0] + 0.3*inputs[1] - 0.7*inputs[2]
neuron.learn_1(inputs, target)
errors.append(neuron.last_error)
print report(errors)
class ProgramLogic:
def __init__(self):
parser.parseFile()
self.neuron = Neuron(self.inputCount())
self.neuron.randomize(-1.0, 1.0)
self.teachingStep = 0
self.prevResponse = 0
self.prevError = 0
self.curResponse = 0
self.curError = 0
def inputCount(self):
return parser.counts[parser.inputCount]
def performTeaching(self, teachingRatio):
resultPrev = self.neuron.learn(self.currentNormalizedInputs(), self.currentExpectedOutput(), teachingRatio)
self.prevResponse = resultPrev[Neuron.prevResponse]
self.prevError = resultPrev[Neuron.prevError]
self.curResponse = self.neuron(self.currentNormalizedInputs())
self.curError = self.currentExpectedOutput - self.curResponse
self.teachingStep += 1
def currentComment(self):
return parser.elements[self.realIndex()][parser.comment]
def currentExpectedOutput(self):
return parser.elements[self.realIndex()][parser.expectedOutputs][0]
def currentInputs(self):
return parser.elements[self.realIndex()][parser.inputs]
def currentNormalizedInputs(self):
return Neuron.normalize(self.currentInputs())
def currentPrevWeights(self):
return self.neuron.weights
def currentPrevResponse(self):
return self.prevResponse
def currentPrevError(self):
return self.prevError
def currentResponse(self):
return self.curResponse
def currentError(self):
return self.curError
def realIndex(self):
return self.teachingStep % len(parser.inputs)
def test_5(steps):
weights = 40
print "Target sqrt(avg) {0}, {1} steps".format(weights, steps)
neuron = Neuron(weights, sigm, sigmp, error)
errors = []
for i in range(steps):
inputs = [random.random() for r in range(weights)]
avg = sum(inputs)/len(inputs)
target = math.sqrt(avg)
neuron.learn_1(inputs, target)
errors.append(neuron.last_error)
print report(errors)
def test_4(steps):
weights = 40
print "Target max - min {0}, {1} steps".format(weights, steps)
neuron = Neuron(weights, sigm, sigmp, error)
errors = []
for i in range(steps):
inputs = [random.random() for r in range(weights)]
imax = max(inputs)
imin = min(inputs)
target = imax - imin
neuron.learn_1(inputs, target)
errors.append(neuron.last_error)
print report(errors)
def __init__(self, size, dt=0.001, tau_rc=0.02, tau_ref=0.002):
""" Constructor for a set of LIF rate neuron
:param int size: number of neurons in set
:param float dt: timestep for neuron update function
:param float tau_rc: the RC time constant
:param float tau_ref: refractory period length (s)
"""
Neuron.__init__(self, size, dt)
self.tau_rc = tau_rc
self.tau_ref = tau_ref
self.voltage = theano.shared(numpy.zeros(size).astype('float32')) # internal variables
self.refractory_time = theano.shared(numpy.zeros(size).astype('float32')) # internal variables
def main(train, test, out):
TRAIN_FILE = train
TEST_FILE = test
OUT_FILE = out
img = Image.open(TRAIN_FILE) # read double moons image from .png file
pixels = img.load() # generate pixel map
width = img.size[0]
height = img.size[1]
training_set = dict()
for i in range(width):
for j in range(height):
if pixels[i,j] == BLUE: # if pixel is blue
training_set[i,j] = BOT # set value to bottom
elif pixels[i,j] == RED: # if pixel is red
training_set[i,j] = TOP # set value to top
# create neuron with 2 input nodes
n = Neuron(2) # x-input, y-input
print "Neuron created."
# training
print "Training..."
counter = 0
while True:
errors = 0
for p in training_set:
errors += n.train_step(p, training_set[p])
counter += 1
print "====="
if errors < n.get_margin() * len(training_set):
break
print "Length of training set: " + str(len(training_set))
print "Iterations: " + str(counter)
# test cases
img = Image.open(TEST_FILE)
pixels = img.load()
width = img.size[0]
height = img.size[1]
for i in range(width):
for j in range(height):
if pixels[i,j] == BLACK:
n.set_input(0, i)
n.set_input(1, j)
n.activate()
ans = n.get_output()
if ans == TOP:
pixels[i,j] = RED
elif ans == BOT:
pixels[i,j] = BLUE
img.save(OUT_FILE, "PNG")
def add_neuron(self):
# return True
if len(self.neurons) >= MAX_NUM_OF_NEURONS:
return False
new = Neuron()
self.neurons.append(new)
self.alpha_neuron.connect(new)
for i in self.next_layer:
for a in i.neurons:
new.connect(a)
for k in self.prev_layer:
for b in k.neurons:
b.connect(new)
self.history.append([1, `new`])
return True
def __init__(self,inputs, hiddentotal):
self.learning_constant = 0.5
self.InputNeurons = []
for i in xrange(inputs):
self.InputNeurons.append(Neuron())
self.InputNeurons.append(Neuron(1)) #adding bias neuron
self.HiddenNeurons = []
for i in xrange(hiddentotal):
self.HiddenNeurons.append(Neuron())
self.HiddenNeurons.append(Neuron(1)) #adding bias neuron
self.OutputNeuron = Neuron()
#connecting everything
for ind_in,val_in in enumerate(self.InputNeurons):
for ind_hid,val_hid in enumerate(self.HiddenNeurons):
c = Connection(val_in,val_hid)
self.InputNeurons[ind_in].addConnection(c)
self.HiddenNeurons[ind_hid].addConnection(c)
for ind,val in enumerate(self.HiddenNeurons):
c = Connection(val,self.OutputNeuron)
self.HiddenNeurons[ind].addConnection(c)
self.OutputNeuron.addConnection(c)
def _create_neurons(self, neuron_count_vec, weight_counts):
neuron_count_vec_length = len(neuron_count_vec)
for i in range(0, neuron_count_vec_length):
self.neurons.append([])
for j in range(0, neuron_count_vec[i]):
weights = Network.get_initial_neuron_weights(weight_counts[i]+1)
inputs = [0] * (len(weights)-1)
inputs.append(1)
neuron = Neuron(i, j, weights, inputs)
if i == neuron_count_vec_length-1:
neuron.is_output_layer = True
self.neurons[i].append(neuron)
def __init__(self, input_size, lrn_rate=1):
"""'input_size' is the length of the input.
'lrn_rate' is the learning rate.
"""
self.neuron = Neuron([0]*input_size, 0, signal)
self.lrn_rate = lrn_rate
self.fire = self.neuron.fire
def __init__(self):
parser.parseFile()
self.neuron = Neuron(self.inputCount())
self.neuron.randomize(-1.0, 1.0)
self.teachingStep = 0
self.prevResponse = 0
self.prevError = 0
self.curResponse = 0
self.curError = 0
def __init__ (self, map_dimensions, data_dimension, kernel_func, nb_clusters=0, verbose=False):
self._verbose = verbose
self._map_dimensions = []
self._map_dimensions[:] = map_dimensions[:]
self._data_dimension = data_dimension
self._dataset = None
self._kernel = kernel_func
self._nb_clusters = nb_clusters
self._neurons = []
self._clusters = []
for position in generate_positions(map_dimensions):
# Initialize _neurons positions in map
neuron = Neuron(position=position)
# Initialize _neurons coordinates
coords = []
for i in xrange(data_dimension):
coords.append(random.random() * 10)
neuron.set_data(coords)
self._neurons.append(neuron)
def updateResult(self):
for i in xrange(len(self.neuron.weights)):
self.neuron.weights[i] = self.widgetExperiment.tableWidget.cellWidget(i, 1).value()
inputs = [self.widgetExperiment.tableWidget.cellWidget(i, 2).value() for i in xrange(len(self.neuron.weights))]
response = self.neuron.response(inputs)
signalStrength = Neuron.strength(inputs, Neuron.StrenghtNormEuclidean)
memStrength = self.neuron.memoryTraceStrength(Neuron.StrenghtNormEuclidean)
self.widgetExperiment.signalEdit.setText(str(signalStrength))
self.widgetExperiment.memoryEdit.setText(str(memStrength))
self.widgetExperiment.outputEdit.setText(str(response))
class Perceptron(object):
"""Online learning Perceptron.
"""
def __init__(self, input_size, lrn_rate=1, activation=signal):
"""'input_size' is the length of the input.
'lrn_rate' is the learning rate.
"""
self.neuron = Neuron([0]*input_size, 0, activation)
self.lrn_rate = lrn_rate
self.fire = self.neuron.fire
def training(self, inputs_vector, outputs, max_epochs):
"""Not checking if inputs_vector and outputs have the same size.
"""
epochs = 0
while True:
epochs = epochs + 1
error_count = 0
for (inputs, output) in zip(inputs_vector, outputs):
actual_output = self.fire(inputs)
error = output - actual_output
if error != 0:
learned = self.lrn_rate*error
self.neuron.update(inputs, learned)
error_count = error_count + 1
if error_count == 0:
break
elif max_epochs and (epochs > max_epochs):
return False
return epochs
def __str__(self):
ret = 'lrn_rate: %s' % self.lrn_rate
ret = '%s\n%s' % (ret, self.neuron.__str__())
return ret
