readme

README.md

@@ -1,2 +1,10 @@
 # PT21-22-Reseau-Neurones
 
+Cette librairie python permet de créer simplement un modèle de perceptron multicouche en choisissant les hyperparamètres suivants:
+ - Le nombre de couches
+ - Le nombre de neurones pour chaque couche
+ - Le learning rate
+ - Le batch size
+ - Le nombre d’époques
+
+ Il est également possible de donner un jeu de données supplémentaire afin de mesurer la précision du modèle à la fin de l'entraînement. Lorsqu’il n'y a que deux neurones d'entrées, il y a une option pour visualiser l’entraînement.

sobek/activationFunctions.py

@@ -1,23 +0,0 @@
-class activationFunction:
-    def applyTo(value):
-        pass
-
-    def applyDerivateTo(value):
-        pass
-
-class sigmoid(activationFunction):
-    def applyTo(value):
-        return 1.0/(1.0+np.exp(-value))
-
-    def applyDerivateTo(value):
-        return sigmoid.applyTo(value) * (1 - sigmoid.applyTo(value))
-
-class reLu(activationFunction):
-    def applyTo(value):
-        return max(0, value)
-
-    def applyDerivateTo(value):
-        return 0 if (value < 0) else 1
-
-class softMax(activationFunction):
-    pass

sobek/layers.py

@@ -1,30 +0,0 @@
-import numpy as np
-
-class layer:
-    def __init__(self, neurons, activationFunction)
-        self.neurons = neurons
-        self.activationFunction = activationFunction
-
-    def process(_input, __storeValues=False)
-
-class dense(layer):
-    def process(_input, __storeValues=False):
-        
-        _input = np.dot(layerWeights, _input)
-        _input = np.add(_input, layerBias)
-
-        if (__storeValues):
-            self.activation = _input
-
-        _input = self.activationFunction.applyTo(_input)
-
-        if (__storeValues):
-            self.output = _input
-
-        return _input
-
-class convolution(layer):
-    pass
-
-class flatten(layer):
-    pass

sobek/network.py

@@ -1,5 +1,6 @@
 import random
 import numpy as np
+import matplotlib
 import matplotlib.pyplot as plt
 import matplotlib.animation as animation
 import pickle
@@ -93,7 +94,7 @@ class network:
                 vizualisationFrame = np.empty((30, 30))
                 for x in range(30):
                     for y in range(30):
-                        vizualisationFrame[x][y] = self.process(np.array([float(x), float(y)]))
+                        vizualisationFrame[x][y] = self.process(np.array([float(x)/30, float(y)/30]))
                 vizualisationData.append([graph.imshow(vizualisationFrame, animated=True)])
 
             inputBatches = [inputs[j:j+batchSize] for j in range(0, len(inputs), batchSize)]
@@ -115,7 +116,6 @@ class network:
                     for layerNumber in range(len(errorSumsWeights)-1, -1, -1):
                         for neuronNumber in range(len(errorSumsWeights[layerNumber])):
                             errorSumsBiases[layerNumber][neuronNumber] += self.__Error(layerNumber, neuronNumber)
-                            #eventuellemtn sortir de boucle
                             errorSumsWeights[layerNumber][neuronNumber] = np.dot(errorSumsBiases[layerNumber][neuronNumber],self.outputs[layerNumber])
 
                 total = 0
@@ -136,7 +136,7 @@ class network:
                 print(self.accuracy(accuracyInputs, accuracyDesiredOutputs))
 
         if (visualize):
-            ani = animation.ArtistAnimation(fig, vizualisationData, interval=100)
+            ani = animation.ArtistAnimation(fig, vizualisationData, interval=100, repeat_delay=1000)
             plt.show()
 
     def __Error(self, layer, neuron):
@@ -181,17 +181,3 @@ class network:
     def networkFromFile(fileName):
         with open(fileName, "rb") as file:
             return pickle.load(file)
-
-
-
-
-class model:
-
-    def __init__(self, inputWidth, inputHeight, inputChannels):
-        self.inputWidth = inputWidth
-        self.inputHeight = inputHeight
-        self.inputChannels = inputChannels
-        self.layers = []
-
-    def add(layerType, activation):
-        self.layers.add()

tests/MNISTDrawingPrediction.py

@@ -14,7 +14,7 @@ class Sketchpad(tkinter.Canvas):
         self.bind("<B1-Motion>", self.add_line)
         self.PILImage = Image.new("F", (560, 560), 100)
         self.draw = ImageDraw.Draw(self.PILImage)
-        self.MNISTNN = network.networkFromFile("MNISTtest2")
+        self.MNISTNN = network.networkFromFile("MNIST30epoch")
         self.predictionLabel = predictionLabel
 
     def add_line(self, event):

tests/flowerGardenLearningVisualization.py

@@ -0,0 +1,17 @@
+#!/bin/python3
+from sys import path
+path.insert(1, "..")
+from sobek.network import network
+import pickle
+
+with open("flowerGardenData", "rb") as file:
+    data = pickle.load(file)
+
+trainPoints = data[0]
+trainLabels = data[1]
+
+myNetwork = network(2, 16, 1)
+
+learningRate = 3.0
+
+myNetwork.train(trainPoints, trainLabels, learningRate, batchSize=100, epochs=3000, visualize=True)

tests/generateSobekFlowerGarden.py

@@ -0,0 +1,34 @@
+#!/bin/python3
+import random
+import numpy as np
+import math
+import pickle
+
+trainPoints = []
+trainLabels = []
+
+random.seed(1216513)
+
+for i in range(1000):
+    x = random.randint(-50, 50)
+    y = random.randint(-50, 50)
+
+    distance = math.sqrt(x**2 + y**2)
+
+    if (distance < 10 or 20 < distance < 30):
+        trainLabels.append(np.ones(1))
+    else :
+        trainLabels.append(np.zeros(1))
+
+    x = (x+50)/100
+    y = (y+50)/100
+
+    trainPoints.append(np.array([x, y]))
+
+print(trainPoints[1])
+print(trainLabels[1])
+
+data = [trainPoints, trainLabels]
+
+with open("flowerGardenData", "wb") as file:
+    pickle.dump(data, file)