Autoencoder
Creation of pre-trained autoencoder to learn the initial condensed representation of unlabeled datasets. This architecture consists of 3 parts:
Encoder: Compresses the input data from the train-validation-test set into a coded representation which is typically smaller by several orders of magnitude than the input data. Latent Space: This space contains the compressed knowledge representations and is thus the most crucial part of the network. Decoder: A module that helps the network to “decompress” the knowledge representations and reconstruct the data from their coded form. The output is then compared to a ground truth. Imports from time import time import numpy as np import keras.backend as K from keras.layers import Dense, Input, Layer, InputSpec, Conv2D, MaxPooling2D, UpSampling2D, Flatten, Reshape, Conv2DTranspose from keras.models import Model from keras.initializers import VarianceScaling from sklearn.cluster import KMeans from sklearn.cluster import MiniBatchKMeans from sklearn import metrics from sklearn.metrics import accuracy_score import matplotlib.pyplot as plt from sklearn.manifold import TSNE from sklearn.decomposition import PCA Loading the data from keras.datasets import mnist from keras.datasets import fashion_mnist import numpy as np # Chargement et normalisation (entre 0 et 1) des données de la base de données MNIST (x_train, y_train), (x_test, y_test) = mnist.load_data() x_train = x_train.astype('float32') / 255. x_test = x_test.astype('float32') / 255. x_train = np.reshape(x_train, (len(x_train), 784)) x_test = np.reshape(x_test, (len(x_test), 784)) Classic Autoencoder # Dimension de l'entrée input_img = Input(shape=(784,)) # Dimension de l'espace latent : PARAMETRE A TESTER !! latent_dim = 10 # Définition du encodeur x0 = Dense(500, activation='relu')(input_img) x = Dense(200, activation='relu')(x0) encoded = Dense(latent_dim, activation='relu')(x) # Définition du décodeur decoder_input = Input(shape=(latent_dim,)) x = Dense(200, activation='relu')(decoder_input) x1 = Dense(500, activation='relu')(x) decoded = Dense(784, activation='relu')(x1) # Construction d'un modèle séparé pour pouvoir accéder aux décodeur et encodeur encoder = Model(input_img, encoded) decoder = Model(decoder_input, decoded) # Construction du modèle de l'auto-encodeur encoded = encoder(input_img) decoded = decoder(encoded) autoencoder = Model(input_img, decoded) Summary # Autoencodeur autoencoder.compile(optimizer='Adam', loss='mse') autoencoder.summary() print(encoder.summary()) print(decoder.summary()) Training autoencoder.fit(x_train, x_train, epochs=20, batch_size=128, shuffle=True, validation_data=(x_test, x_test)) Evaluation # Encode and decode some digits # Note that we take them from the *test* set encoded_imgs = encoder.predict(x_test) decoded_imgs = decoder.predict(encoded_imgs) Visualization n = 10 # How many digits we will display plt.figure(figsize=(20, 4)) for i in range(n): # Display original ax = plt.subplot(2, n, i + 1) plt.imshow(x_test[i].reshape(28, 28)) plt.gray() ax.get_xaxis().set_visible(False) ax.get_yaxis().set_visible(False) # Display reconstruction ax = plt.subplot(2, n, i + 1 + n) plt.imshow(decoded_imgs[i].reshape(28, 28)) plt.gray() ax.get_xaxis().set_visible(False) ax.get_yaxis().set_visible(False) plt.show() Display # Affichage count=1000 idx = np.random.choice(len(x_test), count) inputs = x_test[idx] coordsAC = encoder.predict(inputs) coordsTSNE = TSNE(n_components=2).fit_transform(inputs.reshape(count, -1)) coordsPCA = PCA(n_components=2).fit_transform(inputs.reshape(count, -1)) classes = y_test[idx] fig, ax = plt.subplots(figsize=(10, 7)) ax.set_title("Espace latent") plt.scatter(coordsAC[:, 0], coordsAC[:, 1], c=classes, cmap="Paired") plt.colorbar() fig2, ax2 = plt.subplots(figsize=(10, 7)) ax2.set_title("ACP sur espace latent") plt.scatter(coordsPCA[:, 0], coordsPCA[:, 1], c=classes, cmap="Paired") plt.colorbar() fig3, ax3 = plt.subplots(figsize=(10, 7)) ax3.set_title("tSNE sur espace latent") plt.scatter(coordsTSNE[:, 0], coordsTSNE[:, 1], c=classes, cmap="Paired") plt.colorbar()
