def learnUtil(self,input,target,learningRate=0.01):
"""
This method will learn the weights for the different layers.
"""
delta=[]
lnCases=input.shape[0]
# Run the newtwork
self.run(input)
#Calulate the error factor and update the weights
for index in reversed(range(self.layerCount)):
if index==self.layerCount-1:
# The last layer just before the output
output_delta=self._layerOutput[index]-target.T
squareErr=output_delta**2
error=np.sum(squareErr)
delta.append(output_delta*utils.sigm(self._layerInput[index],True))
else:
# For other layers
output_delta=self.weights[index+1].T.dot(delta[-1])
delta.append(output_delta[:-1,:]*utils.sigm(self._layerInput[index],True))
# Compute Weight Vector
for index in range(self.layerCount):
delta_index=self.layerCount-1-index
if index==0:
layerOutput=np.vstack([input.T,np.ones([1,lnCases])])
else:
layerOutput=np.vstack([self._layerOutput[index-1],np.ones([1,self._layerOutput[index-1].shape[1]])])
weightedDelta=np.sum(layerOutput[None,:,:].transpose(2,0,1)*delta[delta_index][None,:,:].transpose(2,1,0),axis=0)
self.weights[index]-=learningRate*weightedDelta
return error
def run(self, input):
"""
This method will run the network for given input and return the output.
"""
lnCases = input.shape[0]
#Clear out the previous run network data
self._layerInput = []
self._layerOutput = []
#Run the network
for index in range(self.layerCount):
if index == 0:
#Input
layerInput = self.weights[0].dot(
np.vstack([input.T, np.ones([1, lnCases])]))
else:
#Intermediate layers
layerInput = self.weights[index].dot(
np.vstack([
self._layerOutput[-1],
np.ones([1, self._layerOutput[-1].shape[1]])
]))
self._layerInput.append(layerInput)
self._layerOutput.append(utils.sigm(layerInput))
return self._layerOutput[-1].T
def learnUtil(self, input, target, learningRate=0.01):
"""
This method will learn the weights for the different layers.
"""
delta = []
lnCases = input.shape[0]
# Run the newtwork
self.run(input)
#Calulate the error factor and update the weights
for index in reversed(range(self.layerCount)):
if index == self.layerCount - 1:
# The last layer just before the output
output_delta = self._layerOutput[index] - target.T
squareErr = output_delta**2
error = np.sum(squareErr)
delta.append(output_delta *
utils.sigm(self._layerInput[index], True))
else:
# For other layers
output_delta = self.weights[index + 1].T.dot(delta[-1])
delta.append(output_delta[:-1, :] *
utils.sigm(self._layerInput[index], True))
# Compute Weight Vector
for index in range(self.layerCount):
delta_index = self.layerCount - 1 - index
if index == 0:
layerOutput = np.vstack([input.T, np.ones([1, lnCases])])
else:
layerOutput = np.vstack([
self._layerOutput[index - 1],
np.ones([1, self._layerOutput[index - 1].shape[1]])
])
weightedDelta = np.sum(
layerOutput[None, :, :].transpose(2, 0, 1) *
delta[delta_index][None, :, :].transpose(2, 1, 0),
axis=0)
self.weights[index] -= learningRate * weightedDelta
return error
def run(self,input):
"""
This method will run the network for given input and return the output.
"""
lnCases=input.shape[0]
#Clear out the previous run network data
self._layerInput=[]
self._layerOutput=[]
#Run the network
for index in range(self.layerCount):
if index==0:
#Input
layerInput=self.weights[0].dot(np.vstack([input.T,np.ones([1,lnCases])]))
else:
#Intermediate layers
layerInput=self.weights[index].dot(np.vstack([self._layerOutput[-1],np.ones([1,self._layerOutput[-1].shape[1]])]))
self._layerInput.append(layerInput)
self._layerOutput.append(utils.sigm(layerInput))
return self._layerOutput[-1].T
