blusson/PT21-22-Reseau-Neurones
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

plus d'erreurs dans les maths, mais ca ne converge toujours pas

Browse Source
This commit is contained in:
eynard 2021-12-16 23:05:27 +01:00
parent 8189a03abf
commit 619f4762ef
2 changed files with 39 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
sobek/network.py
View File

@ -47,7 +47,7 @@ class network:
if (__storeValues): if (__storeValues):
self.activations.append(_input.copy()) self.activations.append(_input.copy())
#reLu application #activation function application
for neuron in range(len(_input)): for neuron in range(len(_input)):
_input[neuron] = network.__sigmoid(_input[neuron]) _input[neuron] = network.__sigmoid(_input[neuron])
@ -67,8 +67,12 @@ class network:
self.__errors = [[0]*(len(layer)) for layer in self.__weights] self.__errors = [[0]*(len(layer)) for layer in self.__weights]
for _input, desiredOutput in zip(inputs, desiredOutputs): for _input, desiredOutput in zip(inputs, desiredOutputs):
#rempli self.activations et self.outputs
self.__output = self.process(_input, True) self.__output = self.process(_input, True)
self.__desiredOutput = desiredOutput self.__desiredOutput = desiredOutput
for layerNumber in range(len(errorSums)-1, -1, -1): for layerNumber in range(len(errorSums)-1, -1, -1):
for neuronNumber in range(len(errorSums[layerNumber])): for neuronNumber in range(len(errorSums[layerNumber])):
for weightNumber in range(len(errorSums[layerNumber][neuronNumber])): for weightNumber in range(len(errorSums[layerNumber][neuronNumber])):
@ -77,12 +81,17 @@ class network:
total = 0 total = 0
for i in range(len(errorSums)): for layerNumber in range(len(errorSums)):
for j in range(len(errorSums[i])): for neuronNumber in range(len(errorSums[layerNumber])):
for k in range(len(errorSums[i][j])): for weightNumber in range(len(errorSums[layerNumber][neuronNumber])):
errorSums[i][j][k] = errorSums[i][j][k] / len(inputs)
total += errorSums[i][j][k] #Probablement faisable avec une multiplication de matrices
self.__weights[i][j][k] -= learningRate * errorSums[i][j][k] errorSums[layerNumber][neuronNumber][weightNumber] = errorSums[layerNumber][neuronNumber][weightNumber] / len(inputs)
total += errorSums[layerNumber][neuronNumber][weightNumber]
#Probablement faisable avec une somme de matrices
self.__weights[layerNumber][neuronNumber][weightNumber] -= learningRate * errorSums[layerNumber][neuronNumber][weightNumber]
print("Error : " + str(total)) print("Error : " + str(total))
@ -96,9 +105,10 @@ class network:
def __ErrorHiddenLayer(self, layer, neuron): def __ErrorHiddenLayer(self, layer, neuron):
upperLayerLinksSum = 0 upperLayerLinksSum = 0
for upperLayerNeuron in range(len(self.__weights[layer+1]-1)): #Probablement faisable avec une multiplication de matrices
for upperLayerNeuron in range(len(self.__weights[layer+1])):
#A comparer avec un acces direct au erreurs precalcules #A comparer avec un acces direct au erreurs precalcules
upperLayerLinksSum += self.__weights[layer+1][upperLayerNeuron][neuron] * self.__Error(layer+1, neuron) upperLayerLinksSum += self.__weights[layer+1][upperLayerNeuron][neuron] * self.__Error(layer+1, upperLayerNeuron)
return network.__sigmoid(self.activations[layer][neuron], True) * upperLayerLinksSum return network.__sigmoid(self.activations[layer][neuron], True) * upperLayerLinksSum
def __partialDerivative(self, layer, neuron, weight): def __partialDerivative(self, layer, neuron, weight):

28
testLearning.py
View File

@ -4,25 +4,37 @@ from sobek.network import network
random.seed() random.seed()
myNetwork = network(1, 8, 8, 10) myNetwork = network(1, 10)
for j in range(3000): learningRate = 1
for j in range(100000):
inputs = [] inputs = []
desiredOutputs = [] desiredOutputs = []
if (j%50 == 0): if (j%50 == 0):
print(j) print(j)
for i in range(200): for i in range(1000):
inputs.append([random.randrange(10)]) inputs.append([random.randrange(10)])
inputs = np.array(inputs, dtype=object) inputs = np.array(inputs, dtype=object)
for i in range(200): for i in range(1000):
desiredOutputs.append([0]*10) desiredOutputs.append([0]*10)
desiredOutputs[i][9 - inputs[i][0]] = 1 desiredOutputs[i][9 - inputs[i][0]] = 1.0
desiredOutputs = np.array(desiredOutputs, dtype=object) desiredOutputs = np.array(desiredOutputs, dtype=object)
if (j%10000 == 0):
learningRate*= 0.1
myNetwork.train(inputs, desiredOutputs, learningRate)
myNetwork.train(inputs, desiredOutputs, 0.01) print(myNetwork.process(np.array([0.0], dtype=object)))
print(myNetwork.process(np.array([1.0], dtype=object)))
print(myNetwork.process(np.array([2.0], dtype=object)))
print(myNetwork.process(np.array([3.0], dtype=object)))
print(myNetwork.process(np.array([4.0], dtype=object)))
print(myNetwork.process(np.array([5.0], dtype=object)))
print(myNetwork.process(np.array([6.0], dtype=object)))
print(myNetwork.process(np.array([7.0], dtype=object)))
print(myNetwork.process(np.array([8.0], dtype=object))) print(myNetwork.process(np.array([8.0], dtype=object)))
print(myNetwork.process(np.array([1.0], dtype=object))) print(myNetwork.process(np.array([9.0], dtype=object)))