def create_hidden_layers(self, network, f):
for i in range(len(network.hidden)):
l = f.readline().strip()
if l:
raise FileFormatException(f.tell())
l = f.readline().strip().split()
if l[0] != 'b' or (l[-1] not in FUNCTIONS.keys() and (l[-1] != 'koh' and l[-1] != 'gros' and l[-1] != 'bp')):
raise FileFormatException(f.tell())
layer = None
neuron = Neuron()
deriv = None
if l[-1] == 'koh':
conf_file = f.readline().strip()
if not conf_file:
raise FileFormatException(f.tell())
layer = Kohonen(conf_file)
network.conf_file = conf_file
func = types.MethodType(liniowa, neuron)
elif l[-1] == 'gros':
layer = Grossberg(network.conf_file)
l = f.readline().strip().split()
func = types.MethodType(FUNCTIONS[l[-1]], neuron)
deriv = types.MethodType(DER_FUNCTIONS[l[-1]], neuron)
elif l[-1] == 'bp':
if network.conf_file == None:
conf_file = f.readline().strip()
network.conf_file = conf_file
layer = Backpropagation(network.conf_file)
l = f.readline().strip().split()
func = types.MethodType(FUNCTIONS[l[-1]], neuron)
deriv = types.MethodType(DER_FUNCTIONS[l[-1]], neuron)
else:
layer = StandardLayer()
func = types.MethodType(FUNCTIONS[l[-1]], neuron)
layer.bias = map(lambda x: x*(-1.0), self.read_weights(l[1:-1]))
network.layers.append(layer)
neurons_count = 0
for j in range(network.hidden[i]):
line = f.readline().strip().split()
if not line:
raise FileFormatException(f.tell())
neuron = Neuron()
if neurons_count == 0:
neurons_count = len(line)
if len(line) != neurons_count:
raise FileFormatException(f.tell())
neuron.f = func
neuron.der = deriv
neuron.weights = self.read_weights(line)
network.layers[-1].neurons.append(neuron)
if isinstance(layer, Backpropagation):
layer.prev_change = [[0.0 for w in n.weights] for n in layer.neurons]
layer.prev_bias_change = [0.0 for b in layer.bias]
return network
def create_output_layer(self, network, f):
l = f.readline().strip()
if l not in FUNCTIONS.keys() and (l != 'koh' and l != 'gros' and l!= 'bp'):
raise FileFormatException(f.tell())
layer = None
neuron = Neuron()
deriv = None
if l == 'koh':
conf_file = f.readline().strip()
if not conf_file:
raise FileFormatException(f.tell())
layer = Kohonen(conf_file)
network.conf_file = conf_file
func = types.MethodType(liniowa, neuron)
elif l == 'gros':
layer = Grossberg(network.conf_file)
l = f.readline().strip().split()
func = types.MethodType(FUNCTIONS[l[-1]], neuron)
deriv = types.MethodType(DER_FUNCTIONS[l[-1]], neuron)
elif l == 'bp':
layer = Backpropagation(network.conf_file)
l = f.readline().strip().split()
func = types.MethodType(FUNCTIONS[l[-1]], neuron)
deriv = types.MethodType(DER_FUNCTIONS[l[-1]], neuron)
else:
layer = StandardLayer()
func = types.MethodType(FUNCTIONS[l], neuron)
network.layers.append(layer)
for i in range(network.outputs):
neuron = Neuron()
neuron.f = func
neuron.der = deriv
network.layers[-1].neurons.append(neuron)
if isinstance(layer, Backpropagation):
layer.prev_change = [[0.0 for w in n.weights] for n in layer.neurons]
layer.prev_bias_change = [0.0 for b in layer.bias]
return network
