def __init__(self, netWork_id = None, *argin):
NetWork.__init__(self)
self.__id = netWork_id
self.initialize(argin)
self.__synapse[ self.__connection[i][j] ].weight += learningRate * delta_cur[jn] * self.__neuron[i].output
delta_pre[-1] *= self.__neuron[i].output * (1 - self.__neuron[i].output)
delta_cur = delta_pre
delta_pre = []
def epochBatch(self, inputSet, outputSet, learningRate):
pass
def epochOnLine(self, inputValue, outputValue, learningRate):
if (len(inputValue) != len(self.__inputNeuron) or len(outputValue) != len(self.__outputNeuron)):
raise Exception('number of trainning data is not fit to network')
self.probagate(*inputValue)
error = []
for i in range(len(outputValue)):
error.append(outputValue[i] - self.__neuron[-len(self.__outputNeuron) + i])
self.__backprobagate(error, learningRate)
return sum([0.5 * e * e for e in error])
def train(self, inputSet, outputSet, trainType = 'OnLine', learningRate = 0.1, maxIteration = 200, errorLimit = 0):
for epoch in range(maxIteration):
error = 0
if trainType = 'OnLine':
for i in range(len(inputSet)):
error += epochOnLine(inputSet[i], outputSet[i], learningRate)
error /= len(inputSet)
if error <= errorLimit:
return
NetWork.register(MLP)