Initial commit

Tensorflow/tutoriel5/CIFAR_10_vgg.py

@@ -0,0 +1,164 @@
+import tensorflow as tf
+import numpy as np
+from sklearn.utils import shuffle
+import matplotlib.pyplot as plot
+import cv2
+
+def read_cifar_file(file, images, labels):
+    shift=0
+    f=np.fromfile(file, dtype=np.uint8)
+    while shift!=f.shape[0]:
+        labels.append(np.eye(10)[f[shift]])
+        shift+=1
+        images.append(f[shift:shift+3*32*32].reshape(3, 32, 32).transpose(1, 2, 0)/255)
+        shift+=3*32*32
+
+def convolution(couche_prec, taille_noyau, nbr_noyau):
+    w=tf.Variable(tf.random.truncated_normal(shape=(taille_noyau, taille_noyau, int(couche_prec.get_shape()[-1]), nbr_noyau)))
+    b=np.zeros(nbr_noyau)
+    result=tf.nn.conv2d(couche_prec, w, strides=[1, 1, 1, 1], padding='SAME')+b
+    return result
+        
+def fc(couche_prec, nbr_neurone):
+    w=tf.Variable(tf.random.truncated_normal(shape=(int(couche_prec.get_shape()[-1]), nbr_neurone), dtype=tf.float32))
+    b=tf.Variable(np.zeros(shape=(nbr_neurone)), dtype=tf.float32)
+    result=tf.matmul(couche_prec, w)+b
+    return result
+
+def normalisation(couche_prec):
+    mean, var=tf.nn.moments(couche_prec, [0])
+    scale=tf.Variable(tf.ones(shape=(np.shape(couche_prec)[-1])))
+    beta=tf.Variable(tf.zeros(shape=(np.shape(couche_prec)[-1])))
+    result=tf.nn.batch_normalization(couche_prec, mean, var, beta, scale, 0.001)
+    return result
+
+taille_batch=100
+nbr_entrainement=200
+learning_rate=0.01
+labels=['avion', 'automobile', 'oiseau', 'chat', 'cerf', 'chien', 'grenouille', 'cheval', 'bateau', 'camion']
+
+train_images=[]
+train_labels=[]
+read_cifar_file("cifar-10-batches-bin/data_batch_1.bin", train_images, train_labels)
+read_cifar_file("cifar-10-batches-bin/data_batch_2.bin", train_images, train_labels)
+read_cifar_file("cifar-10-batches-bin/data_batch_3.bin", train_images, train_labels)
+read_cifar_file("cifar-10-batches-bin/data_batch_4.bin", train_images, train_labels)
+read_cifar_file("cifar-10-batches-bin/data_batch_5.bin", train_images, train_labels)
+
+test_images=[]
+test_labels=[]
+read_cifar_file("cifar-10-batches-bin/test_batch.bin", test_images, test_labels)
+    
+ph_images=tf.placeholder(shape=(None, 32, 32, 3), dtype=tf.float32)
+ph_labels=tf.placeholder(shape=(None, 10), dtype=tf.float32)
+
+result=convolution(ph_images, 3, 64)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=convolution(result, 3, 64)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=tf.nn.max_pool(result, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
+
+result=convolution(result, 3, 128)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=convolution(result, 3, 128)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=tf.nn.max_pool(result, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
+
+result=convolution(result, 3, 256)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=convolution(result, 3, 256)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=convolution(result, 3, 256)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=tf.nn.max_pool(result, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
+
+result=convolution(result, 3, 512)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=convolution(result, 3, 512)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=convolution(result, 3, 512)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=tf.nn.max_pool(result, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
+
+result=convolution(result, 3, 512)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=convolution(result, 3, 512)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=convolution(result, 3, 512)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=tf.nn.max_pool(result, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
+result=tf.contrib.layers.flatten(result)
+
+result=fc(result, 512)
+result=normalisation(result)
+result=tf.nn.relu(result)
+result=fc(result, 10)
+socs=tf.nn.softmax(result)
+
+erreur=tf.nn.softmax_cross_entropy_with_logits_v2(labels=ph_labels, logits=result)
+train=tf.train.AdamOptimizer(learning_rate).minimize(erreur)
+precision=tf.reduce_mean(tf.cast(tf.equal(tf.argmax(socs, 1), tf.argmax(ph_labels, 1)), tf.float32))
+
+
+with tf.Session() as s:
+    s.run(tf.global_variables_initializer())
+    tab_train=[]
+    tab_test=[]
+    train_images, train_labels=shuffle(train_images, train_labels)
+    for id_entrainement in np.arange(nbr_entrainement):
+        print("> Entrainement", id_entrainement)
+        for batch in np.arange(0, len(train_images), taille_batch):
+            s.run(train, feed_dict={
+                ph_images: train_images[batch:batch+taille_batch],
+                ph_labels: train_labels[batch:batch+taille_batch],
+            })
+        print("  entrainement OK")
+        tab_precision_train=[]
+        for batch in np.arange(0, len(train_images), taille_batch):
+            p=s.run(precision, feed_dict={
+                ph_images: train_images[batch:batch+taille_batch],
+                ph_labels: train_labels[batch:batch+taille_batch]
+            })
+            tab_precision_train.append(p)
+        print("  train:", np.mean(tab_precision_train))
+        tab_precision_test=[]
+        for batch in np.arange(0, len(test_images), taille_batch):
+            p=s.run(precision, feed_dict={
+                ph_images: test_images[batch:batch+taille_batch],
+                ph_labels: test_labels[batch:batch+taille_batch]
+            })
+            tab_precision_test.append(p)
+        print("  test :", np.mean(tab_precision_test))
+        tab_train.append(1-np.mean(tab_precision_train))
+        tab_test.append(1-np.mean(tab_precision_test))
+    quit()
+    plot.ylim(0, 1)
+    plot.grid()
+    plot.plot(tab_train, label="Train error")
+    plot.plot(tab_test, label="Test error")
+    plot.legend(loc="upper right")
+    plot.show()
+    
+    resulat=s.run(socs, feed_dict={ph_images: test_images[0:taille_batch]})
+    np.set_printoptions(formatter={'float': '{:0.3f}'.format})
+    for image in range(taille_batch):
+        print("image", image)
+        print("sortie du réseau:", resulat[image], np.argmax(resulat[image]), labels[np.argmax(resulat[image])])
+        print("sortie attendue :", test_labels[image], np.argmax(test_labels[image]), labels[np.argmax(test_labels[image])])
+        cv2.imshow('image', test_images[image])
+        if cv2.waitKey()&0xFF==ord('q'):
+            break
+

Tensorflow/tutoriel5/README.md

@@ -0,0 +1,45 @@
+# Tutoriel tensorflow
+
+La vidéo du tutoriel se trouve à l'adresse suivante:
+https://www.youtube.com/watch?v=QaazrkwooFA
+
+N'hésitez pas à faire des commentaires sur youtube
+
+## Test de différentes fonctions sur le programme du tutoriel #2
+
+Les graphiques suivants montrent les différentes courbes d'apprentissage réalisées avec le programme du tutoriel #2 (https://www.youtube.com/watch?v=mUyRdiQRJBI). Seule la fonction d'activation est changée.
+L'apprentissage se fait sur 200 cycles et prend environ 35 minutes avec une GeForce 1080; le temps reste sensiblement le même quelque soit la fonction.
+
+#### Fonction d'activation: sigmoid (tn.nn.sigmoid)
+![graph sigmoid](https://github.com/L42Project/Tutoriels/blob/master/Tensorflow/tutoriel5/images/Figure_sigmoid.png)
+
+#### Fonction d'activation: tangente hyperbolique (tn.nn.tanh)
+![graph tangente hyperbolique](https://github.com/L42Project/Tutoriels/blob/master/Tensorflow/tutoriel5/images/Figure_tanh.png)
+
+#### Fonction d'activation: Unité de Rectification Linéaire (tn.nn.relu)
+![graph Unité de Rectification Linéaire](https://github.com/L42Project/Tutoriels/blob/master/Tensorflow/tutoriel5/images/Figure_relu.png)
+
+#### Fonction d'activation: Leaky Relu (tn.nn.leaky_relu)
+![graph Leaky Relu](https://github.com/L42Project/Tutoriels/blob/master/Tensorflow/tutoriel5/images/Figure_leaku_relu.png)
+
+#### Fonction d'activation: Unité Exponentielle Linéaire (tn.nn.selu)
+![graph Unité Exponentielle Linéaire](https://github.com/L42Project/Tutoriels/blob/master/Tensorflow/tutoriel5/images/Figure_selu.png)
+
+#### Ci-dessous, l'ensemble des fonctions dans un même graphique:
+
+#### Erreur sur la base d entrainement:
+![graph base entrainement](https://github.com/L42Project/Tutoriels/blob/master/Tensorflow/tutoriel5/images/Figure_all.png)
+
+#### Erreur sur la base de test:
+![graph base de test](https://github.com/L42Project/Tutoriels/blob/master/Tensorflow/tutoriel5/images/Figure_all_2.png)
+
+## VGGNet sur CIFAR10
+
+N'oubliez pas de récuperer la base cifar10 (binary version) à l'adresse suivante:
+https://www.cs.toronto.edu/~kriz/cifar.html
+
+L'entrainement prend environ 2h40 sur une GeForce 1080
+
+#### Graph de l'entrainement:
+![graph entrainement](https://github.com/L42Project/Tutoriels/blob/master/Tensorflow/tutoriel5/Figure_cifar_vgg.png)
+