1import numpy as np 2from keras.datasets import mnist 3from keras.layers import Dense, Input 4from keras.models import Model 5 6img_rows, img_cols = 28, 28 7 8(x_train, _), (x_test, _) = mnist.load_data() 9x_train = x_train.astype('float32') / 255. 10x_test = x_test.astype('float32') / 255. 11x_train = x_train.reshape((len(x_train), np.prod(x_train.shape[1:]))) 12x_test = x_test.reshape((len(x_test), np.prod(x_test.shape[1:]))) 13 14input_img = Input(shape=(28 * 28,)) 15encoded = Dense(500, activation='relu')(input_img) 16decoded = Dense(28 * 28, activation='sigmoid')(encoded) 17 18autoencoder = Model(input=input_img, output=decoded) 19autoencoder.compile(optimizer='adadelta', loss='binary_crossentropy') 20autoencoder.summary() 21 22autoencoder.fit(x_train, x_train, 23 nb_epoch=10, batch_size=128, shuffle=True, verbose=2, 24 validation_data=(x_test, x_test)) 25 26 27_________________________________________________________________ 28Layer (type) Output Shape Param # 29================================================================= 30input_1 (InputLayer) (None, 784) 0 31_________________________________________________________________ 32dense_1 (Dense) (None, 500) 392500 33_________________________________________________________________ 34dense_2 (Dense) (None, 784) 392784 35================================================================= 36Total params: 785,284 37Trainable params: 785,284 38Non-trainable params: 0
自编码器能够把输入进行编码自学习，然后再解码 单层自编码逻辑比较简单，首先我们创建了一层全连接层采用relu激活函数作为编码层，再采用一层全连接层使用sigmoid作为解码层。然后构建一个自编码器
autoencoder = Model(input=input_img, output=decoded)
这次使用的优化器是adadelta,这个算法是对 Adagrad 的改进，基本不需要我们去设置学习速率