def _initialize_hidden_layers(self, neural_network):
"""Initializes hidden layers, based on defined number of layers."""
number_of_layers = randint(self.init_minimum_layers, self.layers)
""" -1 here (two lines) """
neurons_per_layer = [
randint(1, self.init_maximum_neurons_per_layer)
for _ in range(number_of_layers - 1)
]
hidden_layers = [[
create_neuron(None,
neural_network.bias,
maximum_bias_connection_weight=self.
maximum_bias_connection_weight)
for _ in range(neurons_per_layer[layer])
] for layer in range(number_of_layers - 1)]
# From Jan: Create hidden layers with one neuron with random activation function each.
# hidden_layers = [[create_neuron(None, neural_network.bias)] for i in range(self.layers - 1)]
# Add final hidden layer with one neuron
activation_function = choice(
list(_NON_LINEAR_ACTIVATION_FUNCTIONS.keys()))
hidden_layers.append([
create_neuron(activation_function,
neural_network.bias,
maximum_bias_connection_weight=self.
maximum_bias_connection_weight)
])
# Returns hidden layers.
return hidden_layers
def _create_final_hidden_neuron(self,
bias,
maximum_bias_connection_weight=None):
""" Creates the final hidden neuron """
activation_function = choice(
list(_NON_LINEAR_ACTIVATION_FUNCTIONS.keys()))
return create_neuron(
activation_function,
bias,
maximum_bias_connection_weight=maximum_bias_connection_weight)
def create_neuron(activation_function=None, bias=None, maximum_bias_connection_weight=1.0): """Creates neuron with defined activation function and bias.""" # If activation function not defined, choose activation function at random. if not activation_function: activation_function = choice(list(_NON_LINEAR_ACTIVATION_FUNCTIONS.keys())) # activation_function = choice(list(_ACTIVATION_FUNCTIONS.keys())) neuron = Neuron(array([]), list(), activation_function) # If is biased, connect to bias with random weight. if bias: Connection(bias, neuron, uniform(-maximum_bias_connection_weight, maximum_bias_connection_weight)) return neuron
def create_cnn_neuron(activation_function=None):
if not activation_function:
activation_function = choice(
list(_NON_LINEAR_ACTIVATION_FUNCTIONS.keys()))
what_layer = np.random.rand()
if what_layer >= 0.5:
neuron = ConvNeuron(np.array([]), list(), activation_function)
else:
neuron = PoolNeuron(np.array([]), list(), activation_function)
return neuron
def create_cnn_neuron(activation_function=None, conv_prob=0.9):
if not activation_function:
activation_function = choice(
list(_NON_LINEAR_ACTIVATION_FUNCTIONS.keys()))
what_layer = np.random.rand()
""" TODO: To Konrad: added this probability for a more flexible code """
if what_layer < conv_prob:
neuron = ConvNeuron(np.array([]), list(), activation_function)
else:
neuron = PoolNeuron(np.array([]), list(), activation_function)
return neuron
def create_network_from_topology(topology):
"""Creates neural network from topology."""
# Create bias.
bias = Sensor(array([1]))
# Creates neurons from remaining items in string.
activation_function = choice(list(_NON_LINEAR_ACTIVATION_FUNCTIONS.keys()))
hidden_layers = [[create_neuron(activation_function, bias) for i in range(j)] for j in topology]
# Create output neuron.
output_neuron = create_neuron('identity', bias)
# Connect nodes in neural network.
if len(hidden_layers) > 1:
for i in range(1, len(hidden_layers)): _connect_nodes(hidden_layers[i - 1], hidden_layers[i])
_connect_nodes(hidden_layers[i], [output_neuron])
else:
_connect_nodes(hidden_layers[0], [output_neuron])
# Create neural network.
neural_network = NeuralNetwork(None, bias, hidden_layers, output_neuron)
# Return neural network.
return neural_network
