def calc_error(self, next_layer):
# print "Calculating the error at layer", self.index
connected_nodes = next_layer.neurons # nodes of next layer
weighted_sum = 0
for i in range(len(self.neurons)): # index of each node in the current layer
current_node = self.neurons[i] # references the current node of the current layer
for node in connected_nodes: # nodes of next layer (reference the weight value of each connection)
weighted_sum += node.weights[i] * node.out_error # weights[i] references the 'i'th node of the connecting layer
if current_node.function == activation.getFunc("sigmoid"):
current_node.out_error = (weighted_sum) * current_node.node_out * (1 - current_node.node_out) # sigmoid
else:
current_node.out_error = (weighted_sum) * (1 - current_node.node_out**2) # tanh
def calc_error(self, expected_output):
# print "Calculating error at the Output layer"
network_error = 0
for i in range(len(self.neurons)):
current_node = self.neurons[i]
current_exp = expected_output[i] # expected output for current node
current_act = self.act_out[i] # the actual output for the current node
if current_node.function == activation.getFunc("sigmoid"):
current_node.out_error = (current_exp - current_act) * current_act * (1 - current_act)
else:
current_node.out_error = (current_exp - current_act) * (1 - current_act**2)
network_error += self.neurons[i].out_error**2 / len(self.neurons)
return network_error
def calc_error(self, next_layer):
# print "Calculating the error at layer", self.index
connected_nodes = next_layer.neurons # nodes of next layer
weighted_sum = 0
for i in range(len(
self.neurons)): # index of each node in the current layer
current_node = self.neurons[
i] # references the current node of the current layer
for node in connected_nodes: # nodes of next layer (reference the weight value of each connection)
weighted_sum += node.weights[
i] * node.out_error # weights[i] references the 'i'th node of the connecting layer
if current_node.function == activation.getFunc("sigmoid"):
current_node.out_error = (
weighted_sum) * current_node.node_out * (
1 - current_node.node_out) # sigmoid
else:
current_node.out_error = (weighted_sum) * (
1 - current_node.node_out**2) # tanh
def calc_error(self, expected_output):
# print "Calculating error at the Output layer"
network_error = 0
for i in range(len(self.neurons)):
current_node = self.neurons[i]
current_exp = expected_output[
i] # expected output for current node
current_act = self.act_out[
i] # the actual output for the current node
if current_node.function == activation.getFunc("sigmoid"):
current_node.out_error = (current_exp - current_act
) * current_act * (1 - current_act)
else:
current_node.out_error = (current_exp -
current_act) * (1 - current_act**2)
network_error += self.neurons[i].out_error**2 / len(self.neurons)
return network_error
def assignActivation(self, func_name):
return activation.getFunc(func_name)
def assignActivation(self, func_name):
return activation.getFunc(func_name)
