merge new stuff

Author: nazaninsbr

hidden_layer.py

@@ -26,9 +26,30 @@ def calcOutputs(self):
 for n in self.neurons:
 self.outputValues.append(n.calc_output())
 
+#
+# set Weight Updated
+#
+# def setWeightsUpdated(self , updatedWeights):
+#	#TODO: testing...
+#	x = len(self.getWeights())
+#	updatedWeightsTemp = updatedWeights[:x]
+#	# if updatedWeights is numpy delete is false
+#	del updatedWeights[:x]
+#	for n in self.neurons:
+#	n.updatedWeights(updatedWeightsTemp)
+#
+#
+#
+
 def getOutput(self):
 return self.outputValues
 
+def getWeights(self):
+li = []
+for n in self.neurons:
+li.extend(n.getWeights())
+return np.asarray(li)
+
 
 def setNewInput(self, newInput):
 #print("len of newInput in hiddenLayer",len(newInput))

input_layer.py

@@ -42,6 +42,27 @@ def calcOutputs(self):
 def getOutput(self):
 return self.outputValues
 
+#
+# set weight updated
+#
+# def setWeightsUpdated(self , updatedWeights):
+#	#TODO: testing...
+#	x = len(self.getWeights())
+#	updatedWeightsTemp = updatedWeights[:x]
+#	# if updatedWeights is numpy delete is false
+#	updatedWeights = np.array([:x])
+#	for n in self.neurons:
+#	n.updatedWeights(updatedWeightsTemp)
+#
+#
+#
+
+def getWeights(self):
+li = []
+for n in self.neurons:
+li.extend(n.getWeights())
+return np.asarray(li)
+
 def setNewInput(self, newInput):
 #print ("newInput in input layer: " , newInput)
 self.input_values = newInput
@@ -61,5 +82,6 @@ def setNewInput(self, newInput):
 n.setNewInput(inputval)
 #print("inputval in input layer: " , inputval)
 self.calcOutputs()
+#print("size of output of set new out in input layer: " , len(self.getOutput()))
 return self.getOutput()
 

neural_network.py

@@ -71,7 +71,10 @@ def calcMaxIndex(self, li):
 
 
 def calcOutput(self, x):
+#print("Here")
+#print(x)
 x = int(x)
+#print("x -> ",x)
 self.output = {0:'A', 1:'B', 2:'C', 3:'D', 4:'E', 5:'F', 6:'G', 7:'H', 8:'I', 9:'J'}[x]
 return self.output
 
@@ -87,6 +90,10 @@ def trainNetwork(self):
 def getOutput(self):
 return self.output
 
+# def getOutForInput(self , x):
+#	if self.input_layer_outputs[x] == 'A':
+#	return 
+
 
 def sgdTrain(self):
 print("In training")
@@ -107,30 +114,54 @@ def sgdTrain(self):
 #print("len of nn_inputs: " , len(self.nn_inputs))
 while(imageInd < len(self.nn_inputs)):
 #print("And Now in Second while:)")
-imageInd = 3*coef + remainder
-#print("this is imageInd: ",imageInd)
+#print("this is imageInd: " , imageInd)
 #print("imageInd: ",self.nn_inputs[imageInd][1])
+#print("sgd size input : " , len(self.inlay.setNewInput(self.nn_inputs[imageInd][1])))
 self.input_layer_outputs = self.inlay.setNewInput(self.nn_inputs[imageInd][1])
+#print("sgd size input out : " , len(self.input_layer_outputs))
 #print ("this is the len of input_layer_outputs -> ",len(self.input_layer_outputs))
 #print ("out input layer: ",self.input_layer_outputs)
 #print("hello sabri:)")
 self.hidden_layer_outputs = self.hidlay.setNewInput(self.input_layer_outputs)
 #print ("out hidden layer: " , self.hidden_layer_outputs)
 self.output_layer_output = self.outlay.setNewInput(self.hidden_layer_outputs)
-print("Expected : {}".format(self.nn_inputs[imageInd][0]))
-print("Got : "+self.calcOutput(self.calcMaxIndex(self.output_layer_output)))
+#print("Expected : {}".format(self.nn_inputs[imageInd][0]))
+#print("Got : "+self.calcOutput(self.calcMaxIndex(self.output_layer_output)))
 l2_error = self.sgdCalcError(self.nn_inputs[imageInd][0], self.calcOutput(self.calcMaxIndex(self.output_layer_output)))
 #print("output layer output -> " , self.calcOutput(self.calcMaxIndex(self.output_layer_output)))
 l2_delta = l2_error*sigmoidDeriv(ord(self.calcOutput(self.calcMaxIndex(self.output_layer_output))))
-syn1 = self.outlay.getWeights()
+#print("max index out layer: " , self.calcMaxIndex(self.output_layer_output))
+#print("sgdtrain out layer: " , self.calcOutput(self.calcMaxIndex(self.output_layer_output)))
+#TODO: Test
+syn2 = self.outlay.getWeights()
+syn1 = self.hidlay.getWeights()
+syn0 = self.inlay.getWeights()
+#print(syn0)
 # print("syn1.T is: " , syn1.T)
-l1_error = np.asarray(l2_delta).dot(syn1.T)
-
+l2 = ord(self.calcOutput(self.calcMaxIndex(self.output_layer_output)))
+l1 = ord(self.calcOutput(self.calcMaxIndex(self.hidden_layer_outputs)))
+#l0 = ord(self.calcOutput(self.calcMaxIndex(self.input_layer_outputs)))
+#print("sgdtrain hidden layer" , self.calcOutput(self.calcMaxIndex(self.hidden_layer_outputs)))
+#print("max index input layer: " , self.calcMaxIndex(self.input_layer_outputs))
+#m = self.calcMaxIndex(self.input_layer_outputs)
+#print("sgdtrain input layer" , self.input_layer_outputs[self.calcMaxIndex(self.input_layer_outputs)])
+#print("calc output input layer: " , self.calcOutput(m))
+l1_error = np.asarray(l2_delta).dot(syn2.T)
+l1_delta = l1_error*sigmoidDeriv(l1)
+# print(l1.T)
+syn2 += np.asarray(l1).T.dot(np.asarray(l2_delta))
+#print("syn2: " , syn2)
+#syn1 += l0.T.dot(np.asarray(l1_delta))
+
 # print(syn1)
+# print(syn2)
 # print(l2_error)
 # print(l2_delta)
+# print(l1_error)
+# print(l1_delta)
 
 coef +=1
+imageInd = 3*coef + remainder
 
 
 
@@ -147,7 +178,9 @@ def sgdCreateNetworkAndTrain(self):
 #print("this is inlay input in sgdCreateNetworkAndTrain",inlay_inputs)
 self.inlay = InputLayer(inlay_inputs, 2)
 #print("self inlay in sgdCreateNetworkAndTrain" , self.inlay)
+#print("size inlay_inputs: ",len(self.inlay.getOutput()))
 self.input_layer_outputs = self.inlay.getOutput()
+#print("size input_layer_outputs: ",len(self.input_layer_outputs))
 #print("input layer out in sgdCreateNetworkAndTrain: " , self.input_layer_outputs)
 self.hidlay = HiddenLayer(HIDDEN_LAYER, self.input_layer_outputs)
 #print("hidden layer out in sgdCreateNetworkAndTrain: " , self.hidlay)

neuron.py

@@ -27,6 +27,12 @@ def calc_output(self):
 self.calc_activation()
 return self.output
 
+def updateWeights(self, updatedWeights):
+#TODO: delete one elements 
+self.weights = np.array([])
+np.copyto(self.weights , updatedWeights)
+print("weightsupdated" , self.weights)
+
 def setNewInput(self, newInput):
 cnt = 0
 #print ("newInput in Neuron: " ,newInput)
@@ -45,6 +51,8 @@ def setNewInput(self, newInput):
 #for i in range (0,newInput.length):
 #cnt +=1
 #print(cnt)
+self.pixel_values = np.array([])
+#print("pixel" , self.pixel_values)
 self.pixel_values = np.asarray(newInput)
 #print(len(newInput))
 #print(self.pixel_values.size)

output_layer.py

@@ -24,15 +24,28 @@ def calcOutputs(self):
 for n in self.neurons:
 self.outputValues.append(n.calc_output())
 
+#
+# set Weight updated
+#
+# def setWeightsUpdated(self , updatedWeights):
+#	#TODO: testing...
+#	x = len(self.getWeights())
+#	updatedWeightsTemp = updatedWeights[:x]
+#	# if updatedWeights is numpy delete is false
+#	del updatedWeights[:x]
+#	for n in self.neurons:
+#	n.updatedWeights(updatedWeightsTemp)
+#
+#
+#
+
 def getOutput(self):
 return self.outputValues
 
 def setNewInput(self,newInput):
 self.input_values = newInput
 for n in self.neurons:
 n.setNewInput(self.input_values)
-
-
 self.calcOutputs()
 return self.getOutput()